Cases A New Best: Fractal Meshify 3 Case Review, Thermal Benchmarks, & NoiseMay 15, 2025Last Updated: 2025-05-15We analyze the Fractal Design Meshify 3’s design, specs, build quality, LEDs, and thermalsThe HighlightsThe Meshify 3 comes with 3 fans and focuses on looksAt its price category, there’s a lot of stiff competitionOverall, we like the caseOriginal MSRP: $140 to $220Release Date: May 2025Table of ContentsAutoTOC Our fully custom 3D Emblem Glasses celebrate our 15th Anniversary! We hand-assemble these on the East Coast in the US with a metal badge, strong adhesive, and high-quality pint glass. They pair excellently with our 3D 'Debug' Drink Coasters. Purchases keep us ad-free and directly support our consumer-focused reviews!IntroToday we’re reviewing the Fractal Meshify 3. It ranges from $140 to $220 depending on the level of “integrated atmospheric lighting” you want, but to be fair to Fractal, it does look good with the angular mesh front panel. The good news upfront with this review is that we actually like the case overall.Editor's note: This was originally published on May 15, 2025 as a video. This content has been adapted to written format for this article and is unchanged from the original publication.CreditsTest Lead, Host, WritingSteve BurkeTesting, WritingPatrick LathanCamera, Video EditingVitalii MakhnovetsCameraTim PhetdaraAndrew ColemanWriting, Web EditingJimmy ThangFractal's has added some interesting new features: an air guide in front of the PSU shroud that genuinely helps GPU thermals in our testing, excellent cable aligner clips, and a bizarre PSU filter that blends seamlessly into the side of the case.This case was going to launch months ago, but was quietly pulled from shelves by Fractal at the last minute because the PCIe slots were too tight. Multiple reviewers, including us, noticed that these PCIe slots were too tight for GPUs to slot properly, an issue we had recorded in our notes for the review and were about to film when Fractal sent an email that it was halting the launch. Fractal manually reworked this, meaning someone went through all the cases and shaved them down.Originally, Fractal would have launched just before the worst of the tariffs, but now is launching into a far more expensive import situation. Fractal says that, for now, its MSRP is unchanged.But we’re here to review the case. We have a 3-hour documentary about the tariffs’ impact.The case comes with 3 fans and clearly has a focus on looks with its lighting and front panel design, but also manages to make a number of mechanical choices that affect cooling.Let’s get into the Meshify 3 review.Meshify 3 Ambience Pro RGB OverviewThis is Fractal's third generation of Meshify cases, starting with the Meshify 3 and the larger Meshify 3 XL. We're reviewing the shallow front-to-back Meshify 3 today, which is the size equivalent of the first generation's Meshify C and the second generation's Meshify 2 Compact (watch our review). The Meshify C (watch our review) was $90 at launch in 2017, followed later by the Meshify 2 (watch our review) at $110.The Meshify 3 is launching from $140-$220 (depending on SKU). It’s not distant from Fractal’s Torrent case. We liked that case. It’s priced at around $190 to $230. The increased Meshify series cost goes towards new features.The flashiest new feature is the "integrated atmospheric lighting" in the top-tier Ambience Pro RGB SKU. Fractal also has committed to proprietary connectors for these cases, which we’re not thrilled about. Fractal additionally has a web tool for the new Adjust Pro fan and RGB hub.The cases are sold at $140 for 3x 140mm fans without LEDs, $160 for 3x 14mm RGB fans using proprietary connectors, and $220 for 3x RGB fans with a lighting control hub and additional case lighting around the front panel, shroud, and power button.Meshify 3 Ambience Pro RGB SpecsMeshify 3 TG3.5"/2.5" mountsUp to 2x HDD or 4x SSDDedicated 2.5" mounts2 (6 total)5.25" drive bays0Expansion slots7Motherboard compatibilityE-ATX (max 277 mm), ATX, mATX, Mini-ITXPower supply typeATXFront interface1x USB 3.2 Gen 2x2 Type-C 20Gbps2x USB 3.0 5Gbps1x Combined Mic/AudioTotal fan mounts6x 120 mm or 5x 140 mm + 1x 120mmFront fan mounts3x 120/140 mm (3x Momentum 140 mm PWM/RGB Included)Top fan mounts2x 120/140 mmRear fan mounts1x 120 mmSide fan mountsN/ABottom fan mountsN/ADust filtersPSUFront radiatorUp to 280/360 mm (max 155 mm width)Top radiatorUp to 280/240 mm (max 148 mm width)Rear radiator120 mm (max 128 mm width)Bottom radiatorN/APSU max length180 mmGPU max length349 mmGPU max height176 mm (7 slot) including cablesCPU cooler max height173 mmCable routing space34.5 mmCable coverYesCable organizers3x Cable Aligners, 3x Internal Straps, 3x Rear External StrapsCaptive thumbscrewsTop Rad. Bracket, PSU BracketTool-less push-to-lockSide Panels, Bottom Filter, Front Panel, Top Panel, PSU BracketCase MaterialSteelLeft side panelTempered GlassRight side panelIndustrial SteelCase dimensions (LxWxH)423 x 229 x 507 mmCase dimensions w/o feet/protrusions/screws423 x 229 x 483 mmNet weight8.0 kgPackage dimensions (LxWxH)596 x 350 x 550 mmGross weight10.3 kgSpecs copied from manufacturer materials, please read review for our own measurements and opinionsThe BuildLet’s get into the build quality and features.Some of the exterior is similar to the Torrent, especially the way the top panel slides off. Unlike the Torrent, the top panel is mesh in the Meshify 3, although only the central portion is actually open due to the thick plastic frame underneath. The front panel pops off from the bottom like the North cases, and the top edge (in the Ambience Pro RGB) has pogo pins to connect the ambient front panel lighting.There are three plastic "cable aligners" that can be snapped over the main cable channel. We found them to be well-placed and surprisingly sturdy even with large bundles of power supply cables wedged behind them. Our only complaint is that they're keyed for no discernible reason other than preventing users from flipping the Fractal logo upside-down. The cable management space behind the motherboard is a uniform depth at just over 3cm across the entire case. There's a massive cutout above the motherboard that runs its entire length, although the cutout at the bottom is undersized and difficult to use when an ATX board is installed. Overall, cable management is above average thanks to the cable aligners and the amount of space, with some velcro straps on the back for exterior cable management. The PSU shroud terminates in a plastic "air guide" angled up towards the GPU. It reminds us a little of the NZXT toilet bowl fan from the 2022 H5 Flow (watch our review), but more so the extremely similar "PSU optional cover plate" from Corsair's old 5000D Airflow (watch our review). The 5000D Airflow's single intake fan wasn't positioned to make that part matter, though, whereas the Meshify 3's intake-heavy configuration will actually benefit.The only discrete dust filter in the case is for the PSU. It ejects from the side, which is always a plus for ease of removal, but it's also built into the leg of the case so that there's no externally visible handle. This is completely unnecessary, but it's one of those small features that shows the care that was put into the case's appearance and great attention to detail.Drives can be mounted to a set of adjustable brackets behind the motherboard tray; Fractal has provided a complicated compatibility matrix, but basically, you can either fit up to 2x 3.5" drives or 4x 2.5" drives. An additional 2x 2.5" drives can be attached directly to the floor of the case underneath the air guide, but if you use those slots, you may want to leave the air guide out to make cables accessible.The front of the case can fit either 3x 120mm fans or 3x 140mm fans. The rails rotate to switch between the two sizes and there are cutouts at specific heights for each individual fan, which means the rails don't obstruct the fans at all, but also there's little freedom to reposition fans or mix and match different sizes. All Meshify 3 SKUs ship with three front fans preinstalled, so this is less of an issue than it would be on a bare chassis. When fans are installed on the rails, the fans and rails can be removed as a single unit by undoing two screws, which is awesome.The removable top bracket is extremely easy to work with and fits radiators up to 280mm with plenty of room. Simultaneously installing a 360mm radiator at the front of the case would technically be possible using standard 30mm thick rads, but without much room left for fittings, fans, or thicker radiators, so we don't recommend it. Fractal notes that the front mount is compatible with 360mm (3x120) radiators of up to 420mm total length with the air guide or 445mm without it. Critically though, this does not mean support for 420mm (3x140) radiators.If any radiators are installed at the front of the case, the cable bar has to be shifted back, which makes it harder to route power cables. In addition, maximum GPU length in the Meshify 3 is already only 349mm, and installing a front radiator directly eats that clearance. We recommend relying on the top radiator slot or moving up to the Meshify 3 XL if you really need space for liquid cooling and especially for longer cards.Along the same lines, Fractal claims compatibility with so-called E-ATX boards up to 277mm wide, but that describes the maximum size that can be installed without completely blocking cable cutouts. Realistically, if you want to install anything even slightly larger than ATX, you should be looking at the Meshify 3 XL instead. Even with a normal ATX board, we shifted the cable bar all the way forward to leave space for the 24-pin power cable.As a finer detail point, we’re happy to see that Fractal has finally compromised by adding screw holes to the plastic tabs on the side panels. If you want to add some extra security while moving your case, you can put screws through the tabs to lock them down.The fans and LEDs are complicated and confusing, so that’ll require a whole section -- but we’ll get into the thermal testing first.Meshify 3 Ambience Pro RGB Thermals Grab a GN15 Large Anti-Static Modmat to celebrate our 15th Anniversary and for a high-quality PC building work surface. The Modmat features useful PC building diagrams and is anti-static conductive. Purchases directly fund our work! (or consider a direct donation or a Patreon contribution!)Fractal's review guide claims that the RGB fans are quieter, lower CFM, but higher static pressure (at max speed) than the non-RGB versions of the same fans. Our thermal results with the stock fans are representative of the RGB case SKUs, though any major physical reasons for performance would still be reflected. The standardized fan test is representative of the Meshify 3 in general.For reasons we’ll explain in the fan control section, we had no way to skip the fan hub and connect the fans directly to the motherboard as we usually would for testing. We validated that the fans reached their rated max speed of 1,800RPM +/- 10% with a laser tachometer. However, the tach signal sent back by the Fractal Adjust Pro hub doesn't line up with standard 4-pole fan behavior: HWiNFO misreports 2295 RPM when the fans are really spinning at just over 1,800 RPM in reality. Fractal's software accounts for this and reports RPM accurately.As for comparison cases, we’ll mostly focus on the Antec C8 ARGB (read our review) and Antec Flux Pro (read our review) as close matches. The closest match in appearance and function may be Fractal's own Torrent Compact RGB. We haven't added the Torrent Compact back to our charts since our testing overhaul, but in our review, it performed similarly to the larger original Torrent. That is in our charts, so we can use that as a rough equivalent.We have plenty of other cases on the charts for your own comparisons. Let’s get into the data.CPU Full Load Thermals - Noise-NormalizedWe’ll start with noise-normalized thermals. We use our hemi-anechoic chamber that we spent $250,000 building for validating the noise levels, which allows us to heavily control the noise floor and get super accurate capture.This is the chart. The Fractal Meshify 3 dominates here, planting itself as the new chart leader and within error of the prior leaders. The Meshify 3's all-core average of 40 degrees Celsius over ambient and 44 degrees for the P-cores is significantly better than previous efforts from the Meshify 2 Compact (watch our review) at 48 degrees average and the Meshify C at 54. Those two cases shared a basic design. The Meshify 3 is more of a departure; Fractal has added an airflow guide to the bottom of the case, the stock fans are higher performance, and all three stock fans are installed in the front of the case. That's an arrangement that's worked well for CPU cooling in the North XL (read our review) and Torrent using our test bench, and evidently it works extremely well in the Meshify 3, which is now the coolest case on the chart. The Torrent was a little warmer than the Meshify 3 at 42 degrees above ambient, while the Meshify 3 is instead tied with the previous best results from the Lian Li Lancool 216 (watch our review) and Lancool 207 (read our review). The Antec Flux Pro is a case that Fractal is likely specifically targeting for competition and, like the Lian Li cases, it's within one degree of the Meshify 3. This test is a weak point for the Antec C8 ARGB, which averaged 50 degrees due to its bottom intake layout.GPU Full Load Thermals - Noise-NormalizedHere’s the chart for GPU thermals when noise-normalized.GPU cooling performance in the Meshify 3 was also strong. The equivalent cases from the previous Meshify generations didn't have particularly strong stock performance, with the Meshify 2 Compact averaging 52 degrees above ambient for the GPU and the Meshify C averaging 57. This is still noise-normalized, so it's not purely a matter of more powerful fans in the Meshify 3, which averaged 43 degrees for the GPU and 46 for the GPU memory. Judging by the performance of the North XL and the Torrent here, it seems possible that the Meshify 3's air guide actually does do something. The North XL and its conventional (although ventilated) full-length PSU shroud averaged 44-45 degrees for the GPU depending on the chosen side panel, while the Torrent with its top-mounted PSU and complete lack of a shroud at the bottom of the case averaged 40 degrees. The Meshify 3 with its shroud and air guide averaged 43 degrees, right in the middle. There are more variables at work here, though, so we need another test to know for sure.As for the other highlighted competitors, the Antec C8 ARGB averaged 42 degrees and the Flux Pro averaged 41. Technically they both beat the Meshify 3's 43 degree average, but they're all among the best performing results on this chart.GPU Full Load Thermals - Full SpeedAs part of our full speed testing, we ran a test pass with a cardboard box extension taped onto the end of the PSU shroud to bring it up flush with the front intake fans. This test is at full speed case fans and controlled CPU and GPU fans.Without this extension, the GPU in the Meshify 3 averaged 41 degrees above ambient versus 45 degrees with it. That's a significant increase with the extension, and it shows that it made sense to shorten the shroud and angle it up like a skate ramp. Adding the extension also raised noise levels from a baseline of 37.5 dBA to 38.7 dBA, which we can only really confidently state is a real difference because of our investments in the hemi-anechoic chamber and our test equipment. This is a real change. Although hardly a detectable difference to the human ear, it is a real one and some people might notice it. Fractal’s new shroud design is superior to a square-edge one both thermally and acoustically.Since we've already seen that the Meshify 3 outperforms its predecessors when noise-normalized, it's no surprise at all that it outperforms them when its 1,800 RPM stock fans are maxed out. 37.5 dBA for the Meshify 3 is definitely louder than the 31-32 dBA levels from the Meshify C and Meshify 2 Compact, but thermal performance isn't even close, with the Meshify C only able to maintain GPU temperatures at 56 degrees above ambient. The C8 ARGB tied the Meshify 3's GPU temperature in this test, which is impressive for the Meshify 3 given that the C8 ARGB has bottom intake fans pointed directly into the GPU. The stock Flux Pro and Torrent remain chart toppers at 38 degrees and 39 degrees respectively thanks to their overkill stock fans; this is also evident from the noise levels, with the Torrent's powerful fans pushing it to a maximum of 42.3 dBA to the stock Meshify 3's 37.5 dBA.GPU Full Load Thermals - Standardized FansStandardized fan testing is next.All of the Meshify 3 SKUs ship with fans, so we're just going to use the standardized fan test as an opportunity to compare directly against the previous Meshify generations when they have the same loadout. The Meshify 3 has three defined 140mm fan mounting locations at the front, so rather than sticking our two 140mm intake fans somewhere in the middle, we were forced to pick either the upper two slots or the lower two.The Meshify C and Meshify 2 Compact both had GPU averages of 47 degrees above ambient, while the Meshify 3 literally bookended the chart at 41 degrees with the fans in the lower slots and 53 degrees in the upper ones. If you alter the stock fan configuration, be sure to at least keep one fan in the bottom slot at the front of the case.CPU Full Load Thermals - Standardized FansOn to CPU thermals while using standardized fans: There wasn't much difference in CPU thermals between the upper and lower slot tests; although the upper slot technically did better with an all-core average of 39 degrees, the lower slot arrangement is preferable due to the GPU thermals we just discussed. The Meshify C and Meshify 2 Compact tied at 41 degrees all-core, not that far from the Meshify 3's 40 degree average.VRM & RAM Full Load Thermals - Noise-NormalizedReturning to the noise-normalized results, the Meshify 3 tied the Flux Pro for the best VRM temperature on the chart at 27 degrees above ambient. That's significantly better than its predecessors at 35 degrees for the Meshify 2 Compact and 39 for the Meshify C. The Torrent averaged 29 degrees and the C8 ARGB with its bottom intake layout averaged 33. The ventilated case roof and airflow definitely help here.The SPD hub (or the RAM temperature average) was better, with the Meshify 3's average of 20 degrees above ambient beating all other stock case results on the chart. The Torrent at 21 degrees and Flux Pro at 22 were also high performers, while the C8 ARGB's more GPU-focused cooling left it with a warmer 24 degree average.Fans and LEDsAnd now we’re back to the topic of fans and LEDs.The Momentum RGB fans connect through daisy-chained USB Type-C connectors. Fractal describes the connector as "Proprietary USB-C," meaning that the connectors are just USB-shaped out of convenience; the fans are NOT USB devices. This part confused us, so here's a breakdown:The regular Meshify 3's fans use regular 4-pin PWM fan connectorsThe Meshify 3 RGB's fans use Fractal proprietary connectors, but they're pre-routed with an adapter that terminates in a regular 4-pin fan connector and a 3-pin ARGB connector, so the fans can be connected to a normal motherboard out-of-the-boxThe Meshify 3 Ambience Pro RGB has the same fans as the other RGB SKU, but they're pre-routed to Fractal's Adjust Pro hub. A "legacy splitter" cable is included for OUTput so that regular ARGB devices can be daisy-chained onto Fractal's devices, but the INput adapter cable from the Meshify 3 RGB is not included. In other words, if you get the Ambience Pro RGB, you're using the hub (or waiting for Fractal to sell an adapter cable separately)The Adjust Pro Hub can be controlled through a web tool or Microsoft's Dynamic Lighting feature. Fractal can get away with using a simple web app because the Pro Hub is just a USB device, so it doesn't need low-level access to control the motherboard's fan and RGB headers. There’s no launch-on-startup bloatware and the settings are saved locally to the hub. Currently, the app claims to be fully compatible with Chrome, Opera, and Edge (but not Firefox). A sustained internet connection isn't required for this. We were glad to see Dynamic Lighting compatibility, even if it's not perfect. For instance, if you enable Dynamic Lighting, it breaks Fractal's tool, and if you haven't activated Windows then you can't disable Dynamic Lighting without editing the registry.Fractal's software has a selection of prefab lighting patterns to show off the gen 2 lighting, but the default Northern Lights pattern may be our favorite. The indirect lighting around the edges of the front panel and across the PSU shroud are mostly striking in a dark room. It's unfortunate that such a big part of the case's appeal is limited to the $220 version.Meshify 3 Ambience Pro RGB Conclusion Visit our Patreon page to contribute a few dollars toward this website's operation (or consider a direct donation or buying something from our GN Store!) Additionally, when you purchase through links to retailers on our site, we may earn a small affiliate commission.We like the Meshify 3 overall, but we're torn on several points. As an immediate positive, as reviewers who don’t normally comment a whole lot about looks, we do think they did well to pull off the LEDs and front panel on the Meshify 3. Fractal has consistently done well to execute on this style.As for the more difficult points: First off, the Meshify C launched at $90 and the Meshify 2 Compact at $110. Those weren't the lowest prices in context. The Meshify C, however, was on the affordable end of the scale even for that time. Given that Fractal picked the Meshify 3 MSRPs well before the new tariffs, we believe there's a good chance they'll go up. The Meshify 3 isn't in the medium-budget tier you might assume based on the history of the series. If you're looking for cases in that vein, Lian Li has options like the Lancool 217 (watch our coverage), while Antec has some equivalents to the Meshify 3 RGB like the C8 ARGB and Flux Pro.That's not to say that the Meshify 3 isn't worth the money, just that it’s embattled on all sides. There are a lot of options in this price category, which is good for consumers. The build quality is on par with what we expect from Fractal based on the last few years' worth of cases like the North (watch our review) and Torrent. It’s built well. Thermal performance was extremely good, and although the specs are different for the non-RGB fans as we've explained, we can assume that the performance is also extremely good for the non-RGB case variants. Based on that, the $140 and $160 options are the most competitive.As for the $220 Ambience Pro RGB, it's basically a $60-$80 upcharge for lighting, but we really like that lighting. That is, however, a big price jump. If you're considering the Ambience Pro RGB with the extra cost for the lighting, you should be comparing it to other flagship showy cases like the HAVN HS420 (read our review) or HYTE Y70 (read our review) (if you can buy one as they can be difficult to find in the US due to the tariffs). You should also look through Fractal's documentation carefully to make sure you like the RGB ecosystem of software and connectors before you commit.You can check out the competition like the Flux Pro, C8, and Corsair’s FRAME 4000D (read our review). We have more reviews coming up, but overall, we like Fractal’s new case.

On an unremarkable Monday last February, Andrej Karpathy fired off a tweet that gave the internet its new favorite buzz-phrase: vibe coding.Within hours, people watched tools like Bolt, v0, and Lovable conjure apps from mock-ups never designed or developed.The internet cheered—speed looks spectacular in demo reels—but more senior groups were quietly wincing as AI began to add technical debt to large codebases at previously impossible rates.Why demo-first AI fails mature teamsHere’s how senior designers, developers, and marketers feel the pain.Designers: Prefab design systems ≠ your design systemToday’s one-click generators choose their own colors, border-radii, and fonts. The instant they collide with a house style they don’t recognize, they hard-code new hex values that overwrite your brand tokens, throw off the grid, and leave designers sending screenshots of mismatched styles to the bug tracker.When you draw a picture, you spend your time carefully crafting the shapes and lines. Great design comes from consistency of intention.Today’s AI designers give you a page full of chicken scratches, an eraser, and a “good luck.”Developers: One-shot codegen ≠ production code.A “fully-working” React page arrives as a single 1,200-line component—no tests, no accessibility tags, no separation of concerns. The diff looks like a CVS receipt.The dev who merges it now basically owns a tarball that will resist every future refactor. Code feels like inheriting a junior’s side project, but the junior in this case is an LLM that never sleeps and can’t be taught.Clean, maintainable code is measured by how many lines you can still cut. Without context of your existing codebase or a lot of fine-tuning, AI’s tendency is to be too verbose.Marketers: Fake data ≠ live experiments.Marketing can now crank out a landing page in fifteen minutes, but every testimonial, price, and CTA is still lorem ipsum.Wiring the page to the CMS—and to analytics—means rewriting half the markup by hand. Every 10x sprint lands on an engineering backlog.You can generate pages, sure, but you’re gonna have to wait on the revenue.What’s left for us?Call it all the 80/20 hangover: AI generates the first 80% of the job in 20% of the time… then teams grind through the next 80% of the time patching the final 20% of the work.It’s the perfect inversion of healthy collaboration. Humans drown in drudgery and plead with the machine to handle the craft. But here’s the thing: The lesson isn’t that AI is snake oil; it’s that big team problems aren’t solved by fast AI alone.Sure, “Make me a meal-planning app” is a fun toy, but as we’ve seen, speed without fidelity just moves the bottleneck downstream.What we need is to respect the craft already living in your components, tokens, data, and tests. We need AI for grown-ups.The two-sided contract of AI for grown-upsHere’s the proposal: what if we started building AI with competent teams in mind?For that kind of tooling, two mindset shifts need to happen, both from the builders of AI software and the professionals that use it.Builders need to…Respect the stack. Every pixel, prop, token, schema, and test that already lives in a repo becomes the rails, not a suggestion. Professionals should be able to easily predict what AI outputs will look like. Stay in their lane. Instead of inventing Yet-Another-Canvas, the tooling embeds itself in the software the best teams already use: Figma, IDEs, headless CMSes, etc. Professionals work hard to choose their tools, and AI alone is not a compelling enough reason to switch. Expose control. Codegen happens in daylight—via CLI hooks, readable diffs, and human review gates—so senior engineers can keep the quality bar where it belongs. Professionals are smart enough to handle the machine, so long as it shows them what it’s doing.It’s not up to the professional world to keep perfect pace with the runaway train of AI. It’s the tool builders’ responsibility to stop breaking so much stuff.Professionals need to…Feed the context. Put in the work to document processes, map prototypes to real components, hand over design tokens, and write the first round of tests. Give the model a fighting chance at fidelity. Builders will use this to make AI generations far more deterministic.Stay accountable. A merged PR is still a human name on the line for the most intricate of crafts: UX polish, performance budgets, sparkling copy, etc. Builders don’t have to design all-star AI; they can focus on making it consistent at the grunt-work.Recognize AI limits. Designers, developers, and marketers use AI primarily as the translation layer between deeply skillful fields. Builders don’t need to make AI that replaces teams, but rather AI that fosters communication in the handoffs.It’s not up to the Builders to anticipate every team’s exact use case. Instead, they can trust that professionals will rise to the challenge of adapting AI to their needs. How AI for grown-ups works on a given MondayWell, that’s all great: People focus on people things; AI focuses on AI things. But what does it look like in practice?Let’s take the perspective of our three personas again—designers, developers, and marketers—and see what an ideal world looks like for them.Designers, still in FigmaA designer adjusts the corner radius on the hero card and watches the live hand-off panel flag the exact React prop it will touch.A token audit bot pings him only if the value breaks the design system—no more screenshot spreadsheets.Freed from red-lining, he spends the afternoon nudging micro-interactions and pairing with motion design, really making everything sing.Way more time on polish. No more token-drift bugs filed.Developers, still in the IDEA developer pulls the latest “design-to-code” PR: it’s already split into sensible components, test stubs green-lit, and the diff is small enough to review over coffee.She runs the performance benchmark—numbers hold, thanks to preset budgets the generator never crosses.With boilerplate gone, she dives into accessible keyboard flows and edge-case logic that actually moves the product forward.Review cycles on layout bugs drop by half with AI taking over scaffolding.Marketers, still in the CMSA marketer duplicates last week’s landing page variant, swaps headline copy, and clicks “Stage Experiment.”The AI wires analytics, snapshots the control, and opens a PR tagged for growth-team review—no Jira ticket, no dev backlog delay.They schedule the A/B test in the same dashboard and spend their saved hour crafting social hooks for the launch.Campaign velocity doubles, while engineering time spent on “tiny copy changes” shrinks to near-zero.Same tools, less tedium, deeper ownership. Each role feeds context to the system, stays accountable, and lets the AI do the grunt work—that’s the contract in action.Keep your Cursor, keep your ChatGPT—the glue-layer AI just has to play nice with them while sitting squarely in the team’s shared stack. How we’re building toward AI for grown-upsSo, how close are we to that perfect Monday?At Builder, we find ourselves in a unique position to help, and we’ve been pivoting to meet the demand, especially for large teams.A tiny history lessonBuilder launched in 2019 as a headless CMS plus visual editor whose goal was simple: let developers surface their own JavaScript framework components, and then let non-developers arrange those components without touching code.To pull that off, we built our editor on Mitosis, an open-source layer that describes components once and compiles them to whatever JS framework a team already runs.Because of that early direction, three “grown-up parts” of our product were firmly in place before the generative-AI wave arrived:Fully-featured visual editing: Our Webflow-like editor creates space for designers and marketers to tweak components and pages that compile to any JS framework. Deterministic component mapping: Every Figma frame and its design tokens can round-trip with a real, version-controlled component—not a look-alike snippet. A data source with real content: The CMS holds marketing content, sure, but also all the data engineers need for accurate UI states.So when large language models became practical, we didn’t bolt AI onto a blank slate; we layered it onto an editor, a mapper, and a CMS that were already respecting the stack.That head-start is why our current AI features can focus on removing drudgery instead of reinventing fundamentals.What we’ve made so farOur product works as a bunch of incrementally adoptable layers. No big bang rewrites needed.Visual Editor → Code. Think Bolt or v0, but with Webflow-level tweakability. Prompt the AI to draft a page—or paste any URL for an instant, editable clone—then fine-tune any class, token, or breakpoint by hand. Drop in production-ready components from 21st.dev (or your own repo) and ship designs that respect your CSS, not ours.Figma → Visual Editor (and Code). Install the free Figma Visual Copilot plug-in, draw your frame, and click Export. Copilot converts the frame’s exact Auto-Layout geometry into clean, responsive framework code that can drop straight into the repo or open in Builder’s visual editor for tweaks. Designers still hand over a Figma link; developers run one CLI command to scaffold production-ready code—no guessing, no style drift.Repo → Visual Editor → Repo PR [coming soon]. We’ll be announcing a product shortly that allows anyone to import any GitHub repo to the Visual Editor and make changes that get auto-sent as a PR. Marketers and designers can file engineering tickets with the code already in nice shape.Component Mapping. Tell Builder which Figma button matches which code button once, click “Publish,” and you’re set. From then on, every design export that uses that component drops the real one from your repo into the generated diff—props, tokens, everything—so code reviews focus on ideas, not mismatched styles.Builder Publish: It’s not just a patch for lorem ipsum; it’s your entire headless CMS. Wire any page to real-time content and analytics so marketing can run A/B tests without tapping devs—and get the full schema, versioning, and single-source-of-truth perks that come with a modern CMS, all under the same roof.Where we’re still workingWe’re close, not finished. Next on our list:Less manual context. Auto-detect more tokens and components so mapping feels like autocomplete, not data entry. Deeper control. Let power users crack open every AI draft—props, tests, perf budgets—without leaving the editor. Broader design-system support. Shadcn today, your bespoke kit tomorrow. Mapping should be one click, not a weekend.That “perfect Monday” vision is where we’re headed, and our roadmap will get us there one feature at a time.We’re definitely interested in your feedback: What gaps do you still feel between design, code, and marketing, and where could an “AI for grown-ups” save you the most pain?Drop your thoughts in the comments or tag us on our socials (see footer for links).Turning the 80/20 right-side-upIf Software 2.0 taught us that networks can write code, and the Software 3.0 hype reminds us they can write whole interfaces, then the middle road we’ve walked today—call it Software 2.5—insists that learned programs still deserve linting, budgeting, and daylight.Boring constraints create extraordinary freedom.Speed is intoxicating, but alone it flips the 80/20 rule on its head, leaving humans elbow-deep in cleanup. AI for grown-ups reverses that trade.Tool builders promise rails, not rabbit holes. Professionals promise context and accountability. Every mapped component, every token lock, every human review makes the next generation more predictable—and predictability is what compounds into real velocity.We don’t need to be dragged kicking and screaming into the new world. We’re perfectly capable of walking there at our own pace, spending our hours on craft no model can fake. Introducing Visual Copilot: convert Figma designs to high quality code in a single click.Try Visual CopilotGet a demo

Interviews Daredevil – Born Again: Gong Myung Lee – Production VFX Supervisor By Vincent Frei - 15/05/2025 With over two decades of experience in visual effects, Gong Myung Lee started her career in 2003 and went on to work at renowned studios like Mr. X and Method Studios. She became part of the Marvel Studios team in 2022. Her credits include The Finest Hours, Deadpool 2, Triple Frontier, and The Changeling. What is your background? I started in VFX as a Lighting Technical Director and Compositor, which gave me a deep appreciation for how images work—light, depth, and how all the parts come together. From there, I moved into CG supervision and eventually into VFX supervision, working across features, commercials, and episodic projects. I’ve worked both on the vendor side and in production. At facilities, I built and led teams and tackled creative challenges from inside the post pipeline. Shifting to the production side let me be involved earlier—helping shape decisions on set and ensuring VFX supports the storytelling from day one. How did you get involved on this show? I’d always hoped to work with Marvel on the production side, so when Jen Underdahl called me about this opportunity, I said yes immediately. I’ve been a fan of Daredevil for years—his internal conflict, his vulnerability, and the constant pull between justice and vengeance. And the fact that it was shooting entirely in New York made it even more exciting. Daredevil: Born Again has a distinct visual style. How did you collaborate with the showrunner and directors to ensure that the VFX seamlessly fit within their creative vision? From the start, it was clear this wasn’t going to be a typical superhero show. No lasers, no flying, no giant CG creatures. Daredevil is a street-level hero—he gets hurt, he bleeds, and he wrestles with morality and faith. The VFX had to reflect that same grounded humanity. Everything needed to feel physical, real, and restrained. We worked closely with showrunner Dario Scardapane, directors Justin Benson and Aaron Moorhead, and Marvel creative executive Sana Amanat to define a visual language built around subtlety. The VFX weren’t there to impress—they were there to disappear into the world. We also pulled from key comic book imagery that Sana wanted to preserve—Daredevil crouched on a rooftop, his mid-air baton throws, the skyline behind him. Those became touchpoints for our design language. Being involved from preproduction through delivery meant we could plan closely with every department—art, stunts, SFX, camera—so that VFX could step in where needed. We captured as much in camera as possible, using VFX for what couldn’t be achieved safely or practically. That ranged from invisible stitches to CG weapons, digi-doubles, and full digital builds. Because we had multiple directors and DPs, VFX became a continuity backbone—carrying consistent design and tone across episodes. That applied to everything from the sensory visuals and BB Report graphics, to how Daredevil’s baton behaved from one fight to the next. We also tracked the emotional and visual arc of the show. The intensity of violence escalates as the story deepens. In Episode 102, when Daredevil breaks a cop’s arm, we held back—the blood simply soaks through the sweater, staying grounded in realism. By Episode 109, when Fisk crushes Gallo’s head, the moment is fully on camera. That evolution was intentional. To make it all sit inside the frame, we were rigorous with data acquisition. Every set was LiDAR scanned with photogrammetry, and effects-heavy scenes were supported with HDRIs, cyberscans, clean plates, and lens grids. The goal wasn’t just coverage—it was making sure our work belonged inside the photography. Were there specific requests from the directors regarding the way VFX should enhance the realism and grittiness of the series, particularly in terms of lighting, camera movement, or fight choreography? Absolutely. From the get-go, directors Justin Benson and Aaron Moorhead established a clear visual rulebook. Camera language was closely tied to character: Fisk’s scenes leaned into locked-off frames and symmetry, while Murdock’s were more handheld and reactive. Zooms were preferred over push-ins, and dutch angles were generally avoided. Even when VFX took over a shot—like a CG camera move—we stuck to those same principles to preserve the tone. Our DP, Hillary Spera, shot on Alexa 35s using custom-tweaked anamorphic lenses, which created beautiful lens flares and optical distortions. We mapped those lenses and captured a full library of flare elements so that VFX could match them precisely. That look carried through to DI and helped define the show’s distinct texture. Atmosphere—especially fog and wetdowns—was critical to the show’s look. It shaped lighting and grounded the environments. But in shots requiring CG or 2D elements, we often recreated matching FX atmosphere in post to ensure seamless interaction and integration into the plate. Action followed the same grounded approach. Our Action Director and Supervising Stunt Coordinator, Philip Silvera, choreographed fights based on real-world combat logic, but with emotional weight built into the pacing—breath, exhaustion, impact. For digital takeovers, we started with mocap of stunt performances, then refined using video reference to match the actors’ specific movement and intention. We didn’t shy away from gore either. The directive was clear: keep it grounded, but don’t pull punches. We referenced surgical and forensic material to guide the look of blood hits, wounds, and trauma. Practical prosthetics and squibs formed the base, with VFX adding layers—muscle tear, splatter, and dynamic blood flow—to heighten realism without drifting into stylization. Photo credit: Giovanni Rufino How did you choose the various vendors and split the work amongst them? VFX is a team sport—and like any good team, you need to know your players and play to their strengths. Every facility brings something unique, and a big part of our early conversations was about assigning work based on where each vendor excels. It was about setting everyone up for success. The best VFX comes from the artist’s eye and hand. That final 5%—the nuance and polish—is where the work really comes alive. So I looked for teams who not only had the technical chops but understood the tone of the show and cared deeply about the craft. When that alignment happens, the collaboration just works. We planned the vendor split early, ensuring shared assets would work across facilities and that pipelines were compatible. The Third Floor handled previs, techviz, and postviz. LolaVFX supported early look development. RISE took on some of the most technically complex sequences—like the opening oner at Josie’s, digi-double transitions, and full environment builds. Folks VFX delivered rooftop environments, the finale apartment fight, and key continuity-heavy scenes. Ghost VFX handled the subway fight and Daredevil’s tunnel run, while Phosphene developed the sensory language and crafted bespoke moments like the blackout sequence and car comps. Soho VFX and Powerhouse VFX focused on fight choreography, weapon augmentation, and full-CG sequences. Cantina Creative created all in-world monitor graphics and interfaces. Anibrain, SDFX, and Base FX tackled difficult cleanups, split comps, and continuity patches. Dark Red Studios handled the BB Report design. Scanable handled all LiDAR and cyberscans across our sets and cast—crucial for digi-double and environment work. And we also had an internal team dedicated to the more subtle sensory shots. It was a true team effort, and the final result reflects the strength and dedication each vendor brought to the table. Photo credit: Giovanni Rufino The opening sequence is an incredibly complex oner action scene. Can you walk us through the planning and VFX techniques used to stitch everything together seamlessly? This sequence was a massive effort across all departments, and for VFX, it was one of the most complex and rewarding challenges of the season. The scene begins with Daredevil and Bullseye crashing through the window of Josie’s Bar, moves into the back room, up through a stairway, and eventually emerges onto a rooftop. None of those locations existed together physically, so stitching them into one continuous action required close collaboration between stunts, camera, SFX, and VFX from day one. Our stunt team, led by Philip Silvera, provided a detailed stuntviz that broke the choreography into stitchable beats, which became our foundation. Directors Justin Benson and Aaron Moorhead wanted the camera to feel intentional and fluid—no fast whips or chaotic edits—so every transition had to be precise. There was nowhere to hide a bad seam. RISE led the VFX for the sequence, using a blend of techniques—2D wipes with foreground elements, bluescreen set extensions, FX smoke, and CG transitions—but the real connective tissue was in 3D. We LiDAR-scanned all key locations and rebuilt them in CG, allowing us to bridge multiple stages, extend environments, and support digi-double and FX work with accuracy. Stitching those spaces and action beats required careful spatial continuity and character animation. We used motion capture as the base for digital double work, then enhanced it with keyframed animation to preserve choreography and hit those iconic silhouettes that are signature to Daredevil and Bullseye. The camera tilt moment—where we drop down to see Foggy on the street and then tilt back up to Daredevil and Bullseye—was a digital blend using a 2.5D street projection and a full-CG building facade. That allowed us to link two separate physical locations with one fluid camera move. SFX, supervised by our SFX Coordinator Roy Savoy, handled all the glass breaks and practical sparks inside Josie’s, while Props took care of the breakaway items. Together, they gave us strong in-camera interaction points that VFX could build on. We layered in additional CG debris, neon signage, and more sparks to push the action further. The smoke and atmosphere—inside the bar and on the rooftop—were fully CG to allow precise interaction with digital characters, weapons, gore, and FX. Everything—smoke, blood hits, atmosphere, digi-double transitions—had to feel cohesive and continuous. The real achievement was making the stitching and technical complexity disappear. Given the mix of practical stunts and digital enhancements, what were the biggest challenges in maintaining continuity and realism throughout this oner? One of the biggest challenges was maintaining character integrity—especially for Daredevil. Charlie brings such a specific physicality to the role that even under smoke or in silhouette, his movement reads clearly. Every digital double had to mirror that language—his timing, stance, and how he moves through space—so the audience never questioned the switch. RISE used stunt mocap as the foundation for the digital doubles, then layered in character-specific adjustments to preserve that continuity. At the same time, we were building out full-CG environments to bridge shots, which meant matching textures, lighting, geometry, and camera lensing so that transitions were seamless. Weapons were another layer. Daredevil’s baton, Bullseye’s firearms and knives—all were CG. In one sequence, Daredevil is stabbed by 17 blades, some of which he pulls out to fight back. We had to track each knife’s position, contact shadow, suit tear, and blood drip from shot to shot, which required full-body rotomation and progressive damage built into the suit asset. That also extended to the environment—any damage had to be tracked through the sequence. Grounding the action was essential. Even when Daredevil lands from a rooftop with his grappling hook, it had to feel weighty. We added dust bursts, imperfect footfalls, and a bit of camera vibration to keep it visceral. In a sequence this seamless, every movement, every transition, every detail had to hold up. The illusion depended on total cohesion. How did you approach the digital recreation of New York and Hell’s Kitchen to match the series’ darker, more grounded aesthetic? New York is more than a backdrop—it’s part of Daredevil’s identity. So our approach to the digital builds had to feel authentic, responsive, and emotionally aligned with the show’s tone. We LiDAR-scanned full city blocks, rooftops, and adjacent buildings, and captured tiled plates and HDRIs throughout the day to map real lighting conditions. This gave us a strong base to work from, while still allowing flexibility for framing and layout. Folks VFX handled Matt’s rooftop in Episode 103, while RISE built the Josie’s rooftop and surrounding Hell’s Kitchen. Both spaces had to feel true to the city—textured and believable. Even slight changes in building height or placement could shift the tone of the scene, so we were careful to maintain a balance between realism and composition. Atmosphere was a defining element. We added fog, steam, wetdowns, reflections, and other city-layered effects—like flickering windows and distant traffic—to give each environment weight and depth. These elements weren’t just aesthetic—they helped shape the lighting and made our digital spaces feel physically inhabited and consistent with the show’s tone. Were there any key locations that had to be fully CG or heavily augmented with VFX? If so, what was the process of integrating them into the live-action footage? Yes—several locations had to be fully CG or significantly augmented. Matt’s apartment rooftop in Williamsburg, the Hell’s Kitchen rooftop at Josie’s, the fire escape sequence, Hector’s subway fight, Daredevil’s tunnel run, and the sensory bank vault scene all required extensive digital builds. We began by scanning or photographing real-world references, scouted in collaboration with our creative team, Production Designer and DP. Using LiDAR and photogrammetry, we rebuilt those environments in 3D, preserving scale and lighting using on-set HDRIs. Foreground and midground buildings were modeled in high detail, and wherever windows were visible, we added interior structure and depth to maintain a sense of life beyond the facade. In some cases, city regulations or safety concerns limited our ability to fully scan practical locations. For sequences like the subway platform and the tunnel Daredevil runs through in Episode 106, we used partial scans and reference photography to reconstruct and extend those environments in CG—including the train asset and the tunnel’s structural continuity. In Episode 105, the bank vault sensory sequence was shot on stage with only the real vault door present. Everything else—the spatial layout, dimensionality, and lighting—was designed in CG to reflect the show’s established visual language for Daredevil’s heightened perception. Not exactly an environment extension, but for the “blackout” sequences in the finale, we had to heavily modify our plates to feel like a true city-wide outage. That involved blending day-for-night captures with digital relighting in comp. In Times Square, we replaced the practical LED content with FISK delivering his New Year’s Eve speech. And to build tension, we expanded Fisk’s motorcade in Red Hook using CG mocaped taskforce units and Humvees. Whether it was a full CG build or a heavy plate modification, every element was grounded back into the photography through precise matchmove, consistent lighting cues, and environmental reference. Marvel projects often have a mix of on-location and green-screen work. What techniques did you use to ensure a seamless blend between real and digital environments? The key was always grounding the visual effects in something tangible. Whether we were working on a full bluescreen stage or extending a rooftop, the more real-world reference we could gather, the more seamlessly the VFX would sit in the frame. On Daredevil: Born Again, we captured practical plates and tiled reference footage wherever possible—especially throughout New York. That gave us a photographic base to match to. From there, we built every VFX setup on solid acquisition: LiDAR scans, high-res photogrammetry, HDRIs for lighting, and texture capture from all angles. We avoided relying entirely on CG whenever we could. Even in the most complex shots, we tried to include a practical element—whether it was a foreground prop, atmospheric haze, wet-downs or interactive lighting on set. Those physical components gave us something real to build off and helped the audience feel rooted in the scene. For environments, 2.5D extensions were backed by real plates, and full-CG builds were constructed from real-world geometry and surface data. Even on bluescreen stages, we lit to match the look and feel of the corresponding practical scenes to preserve continuity. We also worked closely with DI to carry through the grade and environmental treatment. That final layer ensured the shot didn’t just blend technically—but emotionally felt like it belonged to the same world. Daredevil is known for its brutal, grounded action sequences. How did VFX help enhance the stunt work while keeping it as practical as possible? Our approach was always practical-first. If it could be done in camera—blood hits, breakaways, physical impacts—we shot it that way. Then VFX would step in to enhance what was already there and bring the moment to full fidelity. For example, blood on faces and inside mouths was done practically using SFX makeup. But for continuity or when choreography called for additional impact, we added digital blood sprays, squibs, or drips—carefully tracked to the performance so they felt embedded, not layered on. Our stunt team gave us strong reference across all key fight beats. We combined that with mocap when needed, especially in sequences that transitioned into digital doubles. Having that real-world base allowed us to maintain physicality and character intention throughout. We also captured a library of flares and lens artifacts from our camera package, giving VFX the right visual language to match the look and feel of the photography—especially during close-quarters combat, where blood, smoke, and light moved dynamically through frame. It helped reinforce the tactile, in-camera aesthetic, even in fully augmented shots. Were there any particular fight sequences that required extensive digital double work, and how did you ensure they remained indistinguishable from the real actors? There were definitely moments where digi-doubles became essential—sometimes for safety, sometimes to maintain camera continuity, and sometimes because the physical environment couldn’t support the choreography as scripted. Sequences like Daredevil’s rooftop runs, tunnel chases, fire escape landings, grappling hook jumps, transitions within the oner, and background extensions to enhance Fisk’s task force all required digital double work at varying levels of complexity. In every action scene, our approach was to ground the work in real-world performance. We captured motion for all key stunts and fight beats, and did detailed cyberscans of the main cast and their stunt doubles. That gave us accurate silhouettes, proportions, and suit detail—everything we needed to maintain visual continuity between live-action and CG. While we didn’t use facial capture rigs or volume setups, Charlie Cox gave us highly expressive reference performances. His focused takes gave us insight into the intention behind each movement, which we either projected in 2.5D or matched in animation for the digi-double moments. What really sold it was the integration. We were meticulous about matching HDRI lighting, lensing, atmosphere, and even subtle depth-of-field cues. Every digital element had to feel like it was captured on the same lens, in the same space, at the same time. That’s what kept the illusion seamless—even in full-CG moments. Were there any memorable moments or scenes from the series that you found particularly rewarding or challenging to work on from a visual effects standpoint? One that stands out is the rooftop environment for Matt’s apartment. It was our first major VFX presentation to the directors and creatives, and it set the tone for what our digital New York would become. It needed to feel cinematic, atmospheric, and emotionally grounded. We scanned and photographed rooftops around Williamsburg, then composited those elements into a custom layout tailored for our scenes. But even subtle changes—like shifting a building’s height or shape—could disrupt the familiarity of the skyline. It’s like adjusting features on a face: subtle, but if it’s wrong, you feel it. Getting that balance right, while still supporting our camera moves and blocking, was incredibly rewarding. Another favorite was developing Daredevil’s sensory language, which we crafted in close collaboration with Phosphene. It evolved throughout the season, but the most ambitious moment was what we called the “Grande Sensory” shot—when Daredevil tunes into a specific sound and his world expands perceptually. We shot it using a custom three-camera rig: a spherical zoom lens in the center, flanked by wide primes. As we pushed in while zooming out—a zolly—we stitched the feeds together at the widest focal length into an approximately 230° field of view, with extra image gathering on the sides. Simultaneously, the aspect ratio expands. Once Daredevil locks onto the target sound, we “unstitch” back to the center camera and compress the aspect ratio as we zoom in for a clean finish. The transitions are subtle but emotionally impactful—designed to be felt, not noticed. We also had the “petite sensory” moments—simpler, analog-feeling visual cues that suggest Daredevil is focusing in. These used optical lens tricks, subtle distortions, and flares—many based on practical elements we shot—so even the heightened perception moments felt grounded in the show’s visual language. And finally, the Gallo head-crush—one of the season’s most graphic shots. The practical prosthetic was impressive, but once pressure was applied, it deformed in a way that didn’t mimic human anatomy. We scanned both the actor and prosthetic for a partial digital takeover—replacing areas with CG muscle, bone, blood, and facial details. Folks VFX handled this sequence and absolutely nailed it, especially the vascular bursts in the eyes and the blink just before the fade. Looking back on the project, what aspects of the visual effects are you most proud of? There’s a lot to be proud of, but what stands out most to me is the consistency of tone and craft—how the visual effects supported the grounded, emotional feel of the show without ever pulling focus. Even in sequences that were technically complex—like the rooftop builds, sensory moments, or the opening oner—we always came back to story. The VFX weren’t about spectacle. They were about precision, intention, and integration. I’m also proud of the continuity we maintained across multiple directors and DPs. Visual effects became a steady thread—carrying design language, violence levels, atmosphere, and choreography from one episode to the next. Most of all, I’m proud that the work feels cohesive. Whether it’s a subtle stitch or a full CG shot, it never distracts. It feels lived in—part of the world, not outside of it. How long have you worked on this show? I was on from November 2022 through November 2024, so a full two years. Being involved early really helped—we had time to plan things properly, make smart choices, and stay ahead instead of playing catch-up. What’s the VFX shot count? We delivered 1,380 final shots in the official turnover cut. What is your next project? We’re already shooting Season 2 of Daredevil: Born Again, which has been exciting. Coming back with the same core team means there’s already a rhythm in place—we know how to work together, and that gives us room to push things further. We’re building on what we set up in Season 1, keeping the tone grounded, but also exploring new visual ideas and storytelling challenges. It’s great to be back in this world, with a chance to evolve it in new ways. What are the four movies that gave you the passion for cinema? There are many, but ones that come to mind immediately are The Godfather I & II, The Shining, In the Mood for Love, and Dune: Part One & Part Two. A big thanks for your time. WANT TO KNOW MORE?RISE: Dedicated page about Daredevil: Born Again on RISE website. © Vincent Frei – The Art of VFX – 2025

You can trust VideoGamer. Our team of gaming experts spend hours testing and reviewing the latest games, to ensure you're reading the most comprehensive guide possible. Rest assured, all imagery and advice is unique and original. Check out how we test and review games here NetEase Games has officially kicked off the much-anticipated Galacta’s Gift event in Marvel Rivals, celebrating the game’s one-year anniversary since its Closed Alpha launch in May 2024. Beginning on May 15, 2025, players can dive into this event that grants a complimentary event pass. This pass features seven tiers packed with exclusive rewards, such as Nameplates, Sprays, eight Chrono Shield Cards, and a Costume Coin applicable to any Season 0 hero. Players can dive into daily missions across Quick Play, Competitive, or AI matches to advance through the pass and snag those highly sought-after items, creating a fulfilling journey for both newcomers and seasoned competitors alike. Yet, lurking beneath the festive exterior of this event is a unique element that has gamers buzzing — the event’s distinctive method of engaging players, especially with its captivating “2099” expiry date. Marvel Rivals Galacta’s Gift event will last forever with unique end date On May 15, trusted leaker X0X_LEAK posted on X that the Marvel Rivals Galacta’s Gift event is “set to end in 2099,” which corresponds with the in-game description, implying it is perpetual. This description, which appears in the event’s UI, shows that Galacta’s Gift is a permanent event for players, particularly as part of the new player onboarding process, which lends credence to the leaker’s assertion of an extended or infinite life. Galacta’s Gift screen in Marvel Rivals. Image by VideoGamer. The 2099 reference, maybe a fun tribute to the Marvel 2099 reality, may inflate the timeline while reflecting the event’s long-term availability. The in-game text indicates that Galacta’s Gift is permanent, as opposed to the time-limited Battle Pass. An X post from Miller Ross also confirms that the event is a one-time completion for existing accounts but is available indefinitely for new players. The event’s design allows for continuous access, therefore, a 2099 “end date” is an exaggerated approach to depict its continuing nature. Players must still finish objectives quickly to receive rewards, as the event’s one-time-per-account constraint implies no second chances. Marvel Rivals Platform(s): macOS, PC, PlayStation 5, Xbox Series S, Xbox Series X Genre(s): Fighting, Shooter Subscribe to our newsletters! By subscribing, you agree to our Privacy Policy and may receive occasional deal communications; you can unsubscribe anytime. Share

May 15, 20253 min readNew Google AI Chatbot Tackles Complex Math and ScienceA Google DeepMind system improves chip designs and addresses unsolved math problems but has not been rolled out to researchers outside the companyBy Elizabeth Gibney & Nature magazine DeepMind says that AlphaEvolve has helped to improve the design of AI chips. MF3d/Getty ImagesGoogle DeepMind has used chatbot models to come up with solutions to major problems in mathematics and computer science.The system, called AlphaEvolve, combines the creativity of a large language model (LLM) with algorithms that can scrutinize the model’s suggestions to filter and improve solutions. It was described in a white paper released by the company on 14 May.“The paper is quite spectacular,” says Mario Krenn, who leads the Artificial Scientist Lab at the Max Planck Institute for the Science of Light in Erlangen, Germany. “I think AlphaEvolve is the first successful demonstration of new discoveries based on general-purpose LLMs.”On supporting science journalismIf you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.As well as using the system to discover solutions to open maths problems, DeepMind has already applied the artificial intelligence (AI) technique to its own practical challenges, says Pushmeet Kohli, head of science at the firm in London.AlphaEvolve has helped to improve the design of the company’s next generation of tensor processing units — computing chips developed specially for AI — and has found a way to more efficiently exploit Google’s worldwide computing capacity, saving 0.7% of total resources. “It has had substantial impact,” says Kohli.General-purpose AIMost of the successful applications of AI in science so far — including the protein-designing tool AlphaFold — have involved a learning algorithm that was hand-crafted for its task, says Krenn. But AlphaEvolve is general-purpose, tapping the abilities of LLMs to generate code to solve problems in a wide range of domains.DeepMind describes AlphaEvolve as an ‘agent’, because it involves using interacting AI models. But it targets a different point in the scientific process from many other ‘agentic’ AI science systems, which have been used to review the literature and suggest hypotheses.AlphaEvolve is based on the firm’s Gemini family of LLMs. Each task starts with the user inputting a question, criteria for evaluation and a suggested solution, for which the LLM proposes hundreds or thousands of modifications. An ‘evaluator’ algorithm then assesses the modifications against the metrics for a good solution (for example, in the task of assigning Google’s computing jobs, researchers want to waste fewer resources).On the basis of which solutions are judged to be the best, the LLM suggests fresh ideas and over time the system evolves a population of stronger algorithms, says Matej Balog, an AI scientist at DeepMind who co-led the research. “We explore this diverse set of possibilities of how the problem can be solved,” he says.AlphaEvolve builds on the firm’s FunSearch system, which in 2023 was shown to use a similar evolutionary approach to outdo humans in unsolved problems in maths. Compared with FunSearch, AlphaEvolve can handle much larger pieces of code and tackle more complex algorithms across a wide range of scientific domains, says Balog.DeepMind says that AlphaEvolve has come up with a way to perform a calculation, known as matrix multiplication, that in some cases is faster than the fastest-known method, which was developed by German mathematician Volker Strassen in 1969. Such calculations involve multiplying numbers in grids and are used to train neural networks. Despite being general-purpose, AlphaEvolve outperformed AlphaTensor, an AI tool described by the firm in 2022 and designed specifically for matrix mechanics.The approach could be used to tackle optimization problems, says Krenn, or anywhere in science where there are concrete metrics, or simulations, to evaluate what makes a good solution. This could include designing new microscopes, telescope or even materials, he adds.Narrow applicationsIn mathematics, AlphaEvolve seems to allow significant speed-ups in tackling some problems, says Simon Frieder, a mathematician and AI researcher at the University of Oxford, UK. But it will probably be applied only to the “narrow slice” of tasks that can be presented as problems to be solved through code, he says.Other researchers are reserving judgement about the tool’s usefulness until has been trialled outside DeepMind. “Until the systems have been tested by a broader community, I would stay sceptical and take the reported results with a grain of salt,” says Huan Sun, an AI researcher at the Ohio State University in Columbus. Frieder says he will wait until an open-source version is recreated by researchers, rather than a rely on DeepMind’s proprietary system, which could be withdrawn or changed.Although AlphaEvolve requires less computing power to run than AlphaTensor, it is still too resource-intensive to be made freely available on DeepMind’s servers, says Kohli.But the company hopes that announcing the system will encourage researchers to suggest areas of science in which to apply AlphaEvolve. “We are definitely committed to make sure that the most people in the scientific community get access to it,” says Kohli.This article is reproduced with permission and was first published on May 14, 2025.

Today, MarkTechPost had the pleasure of interviewing Joey Conway from NVIDIA to discuss their exciting work on open-source large language models, including Llama Nemotron Ultra & Parakeet. Highlights from the interview: NVIDIA’s Open Source Powerhouse: Discover how NVIDIA is pushing the boundaries of open-source AI with the release of cutting-edge models like Llama Nemotron Ultra and Parakeet TDT. Llama Nemotron Ultra: Smaller Size, Giant Performance: Learn how NVIDIA achieved on-par performance with models twice the size, enabling deployment on a single GPU node. Explore their innovative FFN fusion technique for significant speedups. Reasoning on Demand: Uncover the unique “reasoning on/off” feature in Llama Nemotron Ultra, offering unprecedented control for production deployments and cost optimization. Revolutionary Speech Recognition with Parakeet TDT: Dive into NVIDIA’s state-of-the-art ASR model that transcribes one hour of audio in one second with only a 6% word error rate – 50 times faster than other open-source alternatives! The “How”: Architectural Innovations: Get insights into the advanced architectures and optimizations behind these models, including FFN fusion, limited context attention, and the Token Duration Transducer (TDT)  Democratizing AI with Open Data: Learn about NVIDIA’s commitment to the open-source community through the release of model weights and massive, high-quality datasets for both language and speech. Future Directions: Get a sneak peek into NVIDIA’s plans for multilingual support, even smaller edge-optimized models, and advancements in real-time streaming for speech recognition. Production-Ready AI: Understand how these models are designed with real-world deployment challenges in mind, focusing on accuracy, efficiency, and cost-effectiveness. Jean-Marc Mommessin: Joey, welcome to Marketechpost! We’re thrilled to have you here and to delve into the impressive open-source models NVIDIA has been releasing. To start, could you please introduce yourself and your role at NVIDIA? Joey Conway: Hi Jean-Marc, it’s great to be here. I’m Joey Conway, and I work in product management for some of the deep learning software at NVIDIA. Our team focuses on large language models like Nemotron and Llama Nemotron, as well as text-to-speech models such as Parakeet. Jean-Marc Mommessin: Wonderful. And you’ve been at NVIDIA for over seven years now, witnessing significant waves of innovation in AI. Let’s talk about your recent release, Llama Nemotron Ultra, a 253 billion parameter model. From what we’ve seen, it delivers performance on par with models like Llama 405B and DeepSeek R1, which are about twice its size. Remarkably, it can run on a single 8x H100 node. What else can you tell us about Llama Nemotron Ultra and what makes it so impressive? Joey Conway: We’re big believers in the open-source community and the fantastic work being done there. With Llama Nemotron, our goal was to build upon the existing foundations, particularly Llama, for which we greatly appreciate Meta’s contributions. We also observed significant progress in reasoning within the open community earlier this year. Inspired by this, we wanted to contribute and see how we could enhance Llama, especially for enterprise use cases. Our focus was primarily on improving reasoning capabilities and agentic tasks like tool calling and chat. We aimed to take the strengths of the open-source community, enhance them, and then contribute those improvements back. Jean-Marc Mommessin: Did you identify specific gaps in existing models that you aimed to address? You mentioned reasoning, but could you provide an example or two of enterprise agentic tasks where you felt there were shortcomings that Llama Nemotron Ultra overcomes? Joey Conway : Yes, I think looking back to the beginning of the year, a key challenge in enterprise deployments was handling complex queries requiring significant thought and reflection. These could be multi-step processes or involve substantial calculations and the use of external tools. At that time, there weren’t many strong open-weight models capable of robust reasoning. The progress we’ve seen in the last few months in this area is very encouraging. Another critical aspect for enterprises is the ability to accurately call APIs and closely follow instructions in user queries. We wanted to ensure that while we focused on improving reasoning, we didn’t compromise these essential production-level capabilities. Furthermore, we often noticed that when both reasoning and instruction following were well-addressed, they typically resided in separate models. Our aim was to simplify this by creating a single model that excels in both. This was the landscape we observed when we started this project around January and February. Jean-Marc Mommessin: That makes perfect sense and aligns with what we’re seeing in the industry as well. Now, let’s dive into the “how.” Your paper mentions FFN fusion as a key optimization. Could you elaborate on this technique, starting with a high-level explanation? Joey Conway: Absolutely. Our focus on optimization stemmed from the realization that deploying state-of-the-art models often requires a significant deployment footprint. We wanted to optimize this to fit within more common GPU setups. We explored various techniques, including our Puzzle neural architecture search. For dense transformer models, particularly those in the Llama family, we discovered a way to reduce or eliminate redundant attention layers. This process aligned the feed-forward network (FFN) layers in a sequence, allowing us to explore fusion methods. Our fundamental goal on the GPU is to maximize parallel execution. Fusing these aligned FFN layers enables greater parallel computation than was previously possible. By removing redundant layers, we found opportunities to essentially merge or fuse the remaining ones. This is a key example of how we tackle the challenges of running these models at scale. Importantly, this technique often yields greater improvements with larger models, which was beneficial for our Ultra model based on Meta’s Llama 3.1 -405B. Jean-Marc Mommessin: And this FFN fusion significantly improves the model’s throughput, achieving notable speedups. If I recall correctly, it’s in the range of 3 to 5x for the Ultra model? Joey Conway: That’s right, the speedups for the Ultra model are in that range. Additionally, by reducing the model’s size in terms of weights, we also lowered its memory footprint. This allowed us to utilize a larger KV cache. For Llama Nemotron Ultra, we could fit it onto a 8x H100 80GB setup, which is quite significant as it fits within common node configurations. So, FFN fusion provided both a substantial compute speedup and a reduction in memory usage, enabling us to handle larger context lengths. These are very exciting outcomes for us. Jean-Marc Mommessin: Let’s switch gears to data curation. AI data is crucial, and your training pipeline seems very sophisticated. You touched on “instruction following” earlier. Could you elaborate on your data curation process and how you ensured high-quality data, especially considering you leveraged other models in the process? Image source: NVIDIA Joey Conway: Transparency and openness were key in our approach. We wanted to share as much as possible about our data, techniques, and tooling so the community could understand and even use it themselves. Our primary goal with data curation was to improve accuracy across several key domains, including reasoning tasks like math and coding, as well as non-reasoning tasks like tool calling, instruction following, and chat. Our strategy involved curating specific datasets to enhance performance in these areas. Within our supervised fine-tuning process, we differentiated between “reasoning on” and “reasoning off” scenarios. For example, in math and coding, we curated data for simple questions that don’t require complex reasoning, as well as more intricate problems that do. This helps the model learn when and how to apply reasoning. A key part of this process was leveraging high-quality models from the community as “experts” in specific domains. For instance, we used DeepSeek R-1 extensively for reasoning-intensive math and coding tasks. For non-reasoning tasks like basic math, coding, chat, and tool calling, we utilized models like Llama and Qwen. Our aim was to blend the best capabilities of these community models into a single model. We’ve also made this curated dataset publicly available on Hugging Face, with around 30 million question-answer pairs. This allows the community to explore, use, and build upon our work. We were also excited to see our partner ServiceNow recently announce their apprehend Nemotron model, which was trained using our dataset to enhance their own reasoning capabilities. Jean-Marc Mommessin: That’s fantastic that you’re sharing the dataset. Given that you used other models to generate some of this data, what kind of quality checks did you implement to ensure the reliability of the training pairs? Joey Conway: Data quality was absolutely paramount. Since we were generating a significant portion of the data using other models, we implemented a rigorous multi-layered quality assurance process. First, for each expert model used to generate data in a specific domain, we would generate multiple candidate responses for the same prompt. Then, we employed a separate set of “critic” models to evaluate these candidates based on correctness, coherence, and adherence to the prompt. Second, we implemented a scoring mechanism. Each generated question-answer pair received a quality score based on the critic model’s evaluation. We set a high threshold, and any pair that didn’t meet this standard was discarded. Third, human review was integrated at various stages. Our team of data scientists and engineers manually inspected samples of the generated data to identify any systematic errors, biases, or instances of hallucination. This human oversight was crucial for catching nuances that automated systems might miss. Fourth, we focused on the diversity of the generated data. We wanted to ensure we weren’t just getting variations of the same types of questions and answers. We implemented strategies to encourage the expert models to generate a broad range of examples within each domain. Finally, after training Llama Nemotron Ultra on this curated data, we conducted extensive evaluations against benchmark datasets and in real-world use cases. This feedback loop helped us further refine our data generation and filtering techniques. So, it was a comprehensive approach involving expert generation, automated criticism and scoring, human review, diversity checks, and rigorous downstream evaluation to ensure the high quality of our training data. Jean-Marc Mommessin: The quality of the synthetic data is so important. Could you elaborate on the stages you take to ensure high accuracy when generating this data? Joey Conway: Absolutely. When doing synthetic data generation, there are a few key stages to ensure high accuracy. The first is the prompts – the seed data and how we prompt the model. The second is the quality of the responses. On the prompting side, we focus on prompting models where we believe they excel. For example, we might use Llama for chat-related prompts but avoid using a non-reasoning model for math. It’s crucial to align the prompts with the core strengths of the model. For vetting the responses, we invest time in both human manual review and automated methods. Going forward, we anticipate increasing our use of verifiers and reward models, similar to what we’ve done on the Reinforcement Learning (RL) side. The reason we’ve open-sourced much of this is that there’s a lot of nuance involved, and we wanted the community to engage with these challenges. Enterprises like ServiceNow have specific goals, and some of our data might be more or less useful to them. By making it available, they can vet it themselves. We also provide tools like classifier models to help categorize content, such as news or sports, allowing users to make informed decisions about the data blends they use for training. Jean-Marc Mommessin: Perfect. Is there anything else you’d like to highlight regarding this pipeline? Joey Conway: Yes, I’d like to touch on the Reinforcement Learning (RL) aspect. Following the supervised fine-tuning stage, where we enhanced core skills, we’ve just begun to explore the potential of RL with Nemotron. We believe this will be a significant area of future development. What’s exciting about RL is that its effectiveness is largely tied to the available compute time. The more time we invest, the better the model becomes at specific tasks. In our RL stages, we’ve developed methods to automate the process of asking the model a question, grading its answer, and providing feedback to allow it to learn and improve. You can see on the slide the domains where we’ve applied this: scientific reasoning, instruction following, and chat. If you look at the leaderboards, you’ll see that even with new models emerging, we’ve maintained a strong position in these areas, largely due to the effectiveness of RL in achieving top-tier accuracy. We’re optimistic that we’ll see more of this in the community, with more discussion and publication of techniques and data. We’ve started sharing some of our work in this area and will have much more to come in the next three to six months. Jean-Marc Mommessin: You mentioned RL and instruction following, which ties back to the beginning of our conversation. It seems like you’ve come full circle here. Joey Conway: Exactly. The exciting aspect here is automating the feedback loop wherever possible. For chat, we published a fine-tuned reward model last fall. Those who followed our work might recall that our Llama Nemotron model topped the chat leaderboards then. This was because the reward model provides an automated way to teach the original model whether its responses are good or bad. It essentially grades responses based on helpfulness, conciseness, verbosity, groundedness, and similar factors. This granular feedback per generated response allows the model to improve significantly, often more so than through supervised fine-tuning alone, which typically involves a few passes without a continuous feedback loop. Similarly, for instruction following, we use a verifier and a dataset to teach the model whether it followed instructions well or needs to try again. We’re eager to expand this approach to more domains. We’ve already published datasets related to coding and math since the release of this model a few weeks ago, and these have become popular on Hugging Face. I anticipate significant growth in this area within the community. Jean-Marc Mommessin: Alright, so one of the big innovations here, and you touched upon it, but I want to emphasize it, is the ability to toggle reasoning on and off via the system prompt. This is quite unique, and I’m sure many will follow suit. Could you expand on the idea behind this, how you see it applying to agents and beyond, its value, and the key challenges in implementing it? Joey Conway: The reasoning on and off capability was a core goal from the outset. We observed that models in the community often excelled in either reasoning or non-reasoning tasks, and we wanted to simplify deployment by having a single model that could handle both. We had to determine the best way to teach the model when to reason and when not to, while also providing enterprises with explicit control, as they often have deeper domain knowledge than we do. The motivation behind this is that reasoning generates significantly more tokens, which can lead to higher latency and cost. While crucial for solving complex problems, it’s not always necessary. We wanted to give enterprises the control to balance accuracy with latency and cost, allowing them to decide when to employ reasoning and when to opt for faster, less computationally intensive responses. Initially, we weren’t sure how to achieve this, as it hadn’t been widely implemented in the community. Our approach in the supervised fine-tuning stage was to explicitly teach the model by presenting the same question with two different answers: one with detailed reasoning and one without. This essentially doubled our dataset for this specific purpose. However, the outcome is a single model where users can simply include “use detailed thinking on” or “use detailed thinking off” in the prompt to control the model’s reasoning process. On the training side, this required more effort to teach the model this distinction. What we have today is essentially a v1, and I expect others will follow this approach. We’re also excited about future developments, such as time or token limits for reasoning and more granular controls. I’m optimistic that we’ll see further breakthroughs in this area within the next six to nine months, as the problem-solving power of reasoning is significant, but it comes with trade-offs that the community will continue to refine. Jean-Marc Mommessin: We all know that the real test comes in production. Production environments are sensitive to latency, cost, and while accuracy and reasoning are vital, excessive reasoning can lead to scalability issues and increased latency. The flexibility you’ve introduced is fantastic, and I can see numerous production use cases that will greatly benefit from the ability to control reasoning on a per-query basis. So, when you were developing this model, you aimed to balance accuracy and efficiency. Could you share some insights into how you made these trade-offs, the timeline for building the model and the team involved, and how you determined the optimal compromise between these two critical factors? Joey Conway: Balancing accuracy and efficiency is always a challenge. Our initial goal was to achieve both, which is a difficult undertaking. We started with the “Super” model, which was the most recent Llama 3.1 70B release from Meta, as our baseline for accuracy. We weren’t sure if we could simultaneously improve accuracy and reduce the model size. We found that through our training techniques and distillation process, we could indeed boost accuracy. We even released an initial checkpoint reflecting this. However, we wanted to go further by incorporating strong reasoning capabilities, aiming for state-of-the-art reasoning scores. This is where the SFT and RL stages came in, which required significant time for synthetic data generation since this type of data didn’t exist. During training, we carefully considered the number of epochs for each skill and continuously measured accuracy. Our goal was to improve performance across all six key areas rather than excelling in just a couple. This balancing act took more time as we experimented to find the right combinations. However, we felt it was crucial to ensure world-class performance in these six enterprise-relevant scenarios, including chat and instruction following. For areas like MMLU, we focused on maintaining performance and preventing regression rather than actively trying to improve scores. So, there were definitely priorities and trade-offs involved. Ultimately, we believe these were the right focus areas for our enterprise customers. Jean-Marc Mommessin: You are releasing this model family as part of the open-source community. We’ve discussed the gaps you aimed to address and the unique reasoning on/off feature for production scalability. Could you share your thoughts on how NVIDIA and your team view the role of these models within the broader open-source and LLM ecosystem, especially given your work building upon the Llama base? Joey Conway: NVIDIA has a long history of contributing models to the open-source community. What excites us about Llama is its strong traction with enterprise customers. While NVIDIA Research publishes extensively across various domains, our goal with Llama Nemotron was to build upon Llama’s momentum in enterprise adoption by focusing narrowly on specific areas. The base Llama models already cover many things exceptionally well, so we saw an opportunity to build on top of that and be very targeted in our enhancements. The recent LlamaCon event and Meta’s announcements sound very promising, and we’re excited about Llama 4 and the ongoing work there. Moving forward, we anticipate continuing to identify specific areas where we can add significant value, while Meta continues to build excellent general-purpose models suitable for enterprise production. From our perspective, reasoning will likely remain a key focus, and we’re also excited about Meta’s advancements in this area. Tool calling, instruction following, and chat are also areas we’ll continue to develop. One area we’re particularly interested in exploring is multilingual capabilities. For large enterprises, supporting multiple languages is crucial. While many models handle individual languages well, we aim to focus on a few key languages and ensure world-class accuracy for reasoning, tool calling, and chat within those. This is likely the next major area of expansion for us, beyond the exciting developments in model architectures like Llama 4’s new MoE architecture, which we’re also keen to explore for potential distillation and optimization for NVIDIA GPUs. So, there’s a lot of exciting work ahead. Jean-Marc Mommessin: When you say multilingual, are you thinking of supporting a broad range, like 50 languages, or a more focused set, perhaps around 5 or 10 initially, given the benchmark challenges you mentioned? Joey Conway: We’ll probably start with a more focused set, perhaps around 5 to 10 languages. The challenge is that the community currently lacks comprehensive benchmarks for tasks like reasoning or tool calling across a wide variety of languages. As we develop these multilingual models, we’re also having to create evaluation data simultaneously, which takes time. If those benchmarks were readily available, the process would be smoother. However, we see this as an exciting challenge. Our initial focus will likely be on a smaller set of languages where we can establish strong performance, given the current limitations in community-wide benchmarks. Jean-Marc Mommessin: Let’s shift gears and talk about another state-of-the-art open-source model you recently released: Parakeet TDT 0.6 B parameters, V2. This model has set a new standard for automatic speech recognition (ASR), transcribing one hour of audio in just one second. That’s 50 times faster than other open-source ASR models, and remarkably, it achieves only a 6% word error rate. This is truly impressive. What else would you like to highlight about this model before we discuss the “how” behind its incredible performance? Joey Conway: It’s worth noting that NVIDIA has been working on ASR models for a long time, even before I joined. We’ve also released many open models in this space over the years. The teams working on this are exceptional, and they consistently strive to balance accuracy with latency and throughput. Parakeet V2 is the latest in this line of high-performance models from NVIDIA. Jean-Marc Mommessin: It sounds like the advancements will keep coming. So, let’s delve into how you achieved this remarkable performance with Parakeet TDT. What kind of architecture did you use? I understand it’s based on a Fast Conformer architecture with specific optimizations like 8x depth-wise separable convolutional downsampling and limited context attention. Could you explain how you arrived at this approach and whether these optimizations primarily enhance speed and throughput or if they also contribute to accuracy and the ability to process long audio segments like a full hour in one shot? Joey Conway: Yes, we’ve explored various architectures for ASR over the years, and the Conformer architecture, originally from Google, has shown great promise. Our goal with Parakeet TDT was to take the Conformer architecture and make it significantly more efficient and faster without sacrificing quality. We’ve implemented several key optimizations.  First, as you mentioned, the depth-wise separable convolution downsampling. At the input stage, we significantly downsample the audio, which reduces the computational cost and memory requirements for processing. Second is the limited context attention. By focusing on smaller, overlapping chunks of audio, we can maintain accuracy while achieving a speedup in processing. Third, on the encoder side, we also utilize a sliding window attention technique, which allows us to process longer audio files without having to split them into shorter segments. This is crucial for handling long-form audio like a full hour in a single pass. Beyond the Conformer architecture, Parakeet TDT incorporates a Token and Duration Transducer (TDT). Traditional Recurrent Neural Network (RNN) transducer technology processes audio frame by frame. What we’ve done with TDT is enable the model to predict both the tokens and the expected duration of those tokens. This allows it to make decisions to skip over redundant frames, significantly speeding up the transcription process. This TDT innovation alone contributes to around a 1.5 to 2x speedup. So, there’s a combination of architectural choices and specific optimizations that contribute to Parakeet TDT’s impressive speed and accuracy. Jean-Marc Mommessin: I want to go back to one or two of those. Those are amazing, frankly. The speed increase is remarkable. Joey Conway: Yes, and we have another technique called a label looping algorithm. Essentially, when we’re doing batch inference, this algorithm allows us to advance the tokens independently for different samples. This separation of the workflow enables us to sweep and loop over frames and labels more efficiently, significantly speeding up the decoding process. Lastly, on the decoder side, we’ve moved some of the computation into CUDA graphs, which is a more efficient way to run many small kernels. This optimization alone provided around a 3x speed boost. So, as you can see with TDT models, we’ve been able to achieve speeds comparable to Connectionist Temporal Classification (CTC) decoders, which are also known for their speed, while maintaining high accuracy. Our overall theme is always to balance speed improvements with maintaining or even enhancing accuracy. Techniques like CTC decoders have been around for a while and are fast but might not be as accurate. It really depends on the use case, but we’re always striving for that balance. Jean-Marc Mommessin: Can we revisit the limited context attention? Do you see this technique having broader applications in other areas down the line? Joey Conway: Yes, I believe so. Patterns like the sliding window attention are already used in other areas, such as LLMs. Our research teams are constantly experimenting, looking at successful techniques from different domains, and trying to apply them in new ways. Interestingly, some of the researchers who worked on Parakeet TDT also work on Llama Nemotron, so there’s a cross-pollination of ideas. I do expect that some of these techniques will find broader applications going forward. We also anticipate further improvements to TDT and the Conformer architecture, as we’ve been working on them for several years now. I don’t see these core technologies going away anytime soon; we’ll likely continue to refine them. Jean-Marc Mommessin: Leaving the TDT aside, do you see other potential applications for the Token and Duration Transducer concept in other domains? Joey Conway: That’s a good question. I’m not immediately seeing a direct application of the TDT concept outside of ASR. Its history is rooted in RNNs and RNN transducers, which have primarily been used in speech recognition. However, some of the underlying techniques we’ve applied to it, like using CUDA graphs for optimizing kernel execution, are general techniques that we use whenever we identify bottlenecks in a model’s pipeline. So, while the TDT itself might be domain-specific, some of the optimization strategies we’ve employed could certainly translate to other areas, including large language models. Jean-Marc Mommessin: let’s talk about data. AI data is always a key topic. How do you ensure that the data used to train Parakeet TDT is diverse enough to handle various accents, dialects, vocal ranges, pitches, and noisy background conditions, which often negatively impact ASR performance? Joey Conway: You’re absolutely right. As humans, we naturally filter out accents and background noise to understand speech. However, deep learning models are only as good as the data they’re trained on. Early on, limited data for specific accents or languages resulted in poor performance for those variations. What might have initially seemed like edge cases have become increasingly common, highlighting the need for more representative data. We’ve invested significant effort in curating our datasets to reflect this real-world diversity. We use techniques like classifiers to analyze our data and understand the distributions of accents, dialects, and acoustic conditions. We’ve worked with customers like YUM! Brands, who have drive-through use cases with significant highway noise, illustrating the importance of training the model to handle such challenging environments. Ensuring the right blend and distribution of these conditions in our training data is crucial for the model’s robustness. I’m also excited to announce that we plan to open-source a substantial speech dataset, around 100,000 hours, where we’ve meticulously performed this kind of curation. This dataset will include variations in sound levels, signal-to-noise ratios, background noise types, and even telephone audio formats relevant for call centers. Our goal is to provide the community with high-quality, diverse data that enables models to perform well across a wide range of real-world scenarios. Jean-Marc Mommessin: That’s fantastic news about the open-sourcing of the speech dataset! My final question regarding the Parakeet family: you currently have the 600 million and 1.1 billion parameter models. How do you envision future development for this family? What are the potential directions? Joey Conway: We’re considering development along two main dimensions: model size and the number of supported languages. In terms of size, we’ve released models at the smaller and mid-range to demonstrate the potential, similar to our approach with Llama Nemotron Super. We plan to explore larger models, potentially around 2 billion parameters, which we anticipate will handle even more languages and dialects. On the smaller end, we’re even considering models down to around 50 million parameters. The motivation here is to address use cases at the edge where a smaller footprint is necessary, such as enabling real-time audio processing for robots in noisy environments. We’ll be exploring the right trade-offs for such applications. Technologically, we plan to work on streaming capabilities for TDT. Currently, much of the processing is done in an offline batch mode, but we want to enable real-time, live transcription. And as mentioned, we’re excited about releasing the large, curated speech dataset. Finally, for those looking to deploy these models in production, we recommend exploring techniques like word boosting, which allows for customization of text normalization to include domain-specific terms and acronyms. We aim to provide a wide range of options for users to get started and tailor the models to their specific needs. Jean-Marc Mommessin: I’m very familiar with the NVIDIA Orin platform. Would these Parakeet models currently run on NVIDIA Orin? Joey Conway: Yes, I believe the 0.6 billion parameter model likely would run on Orin. I would need to double-check the exact specifications, but I’m quite confident it’s feasible. Jean-Marc Mommessin: Orin packs a significant punch. I especially love the robotics use case you mentioned. While there’s been a lot of focus on robot vision, the ability to hear and understand quickly is equally crucial, especially for safety. A model that’s 50 times faster and highly accurate in understanding another modality seems like a perfect fit for robotics. Joey Conway: Yes, and the slight hesitation I had earlier was due to the understanding that in robotics, there are often multiple models running simultaneously, including vision models. So, resource allocation is a consideration. However, our push towards smaller, more efficient models is precisely to address these kinds of multi-modal edge computing scenarios. The low latency and real-time processing capabilities of Parakeet are indeed very beneficial for enabling robots to react quickly and safely to auditory cues. Jean-Marc Mommessin: Anything else you’d like to add as a final thought on the Llama Nemotron Ultra and Parakeet families? They’re both open-source, fast, high-throughput, cost-efficient, and run on smaller footprints – are these the key takeaways? Joey Conway: Yes, that’s a great summary. Those were the core objectives we set out to achieve. We aimed for state-of-the-art accuracy, optimized footprints for efficient GPU utilization in terms of latency and throughput, and a commitment to open-sourcing everything to empower the community. We’ve strived to be as community-friendly as possible by releasing datasets, using permissive licenses, and making it easy for people to experiment. We’re eager to see the community’s feedback and the innovative applications they build upon our work. We’re also looking forward to learning from their experiences. Jean-Marc Mommessin: Where are all these models and datasets available? Joey Conway: Everything we’ve published is on Hugging Face – the models and the datasets. The software stack to run them comes from NVIDIA and is available on NGC, our content repository. Much of the underlying software is also open-source and can be found on GitHub. We also provide pip wheels for easier installation. The Nemo framework is the central hub for much of this software stack, whether you want to run the models or fine-tune them. We’ve tried to make it as user-friendly as possible. We use the same software internally to build the models, so it should be relatively straightforward for others to pick up and deploy as well. Jean-Marc Mommessin: Well, Joey, this has been fantastic. I’m continually impressed by NVIDIA’s commitment to giving back to the community with state-of-the-art models that will undoubtedly find their way into production. Thank you so much for your time and insights. I look forward to our next conversation. Joey Conway: Thank you, Jean-Marc. It was my pleasure, and we appreciate the opportunity.  Jean-marc MommessinJean-marc is a successful AI business executive .He leads and accelerates growth for AI powered solutions and started a computer vision company in 2006. He is a recognized speaker at AI conferences and has an MBA from Stanford.Jean-marc Mommessinhttps://www.marktechpost.com/author/jean-marc0000677/NVIDIA AI Releases HOVER: A Breakthrough AI for Versatile Humanoid Control in RoboticsJean-marc Mommessinhttps://www.marktechpost.com/author/jean-marc0000677/Speech-to-Speech Foundation Models Pave the Way for Seamless Multilingual InteractionsJean-marc Mommessinhttps://www.marktechpost.com/author/jean-marc0000677/Lowe’s Revolutionizes Retail with AI: From Personalized Shopping to Proactive Customer AssistanceJean-marc Mommessinhttps://www.marktechpost.com/author/jean-marc0000677/Google DeepMind’s Gemini Robotics: Unleashing Embodied AI with Zero-Shot Control and Enhanced Spatial Reasoning

How co-development companies are navigating a turbulent industry in 2025 GamesIndustry.biz catches up with some of the top companies working in the co-dev space to see how they are faring given current industry headwinds Image credit: Bethesda Softworks Feature by Alex Calvin Contributor Published on May 15, 2025 In one form or another, co-development companies have nearly always been around in the games industry. For as long as people have been making games, there have been companies and developers who have made a living helping out other studios with their projects. For the most part, these companies had a low profile, sitting in the shadows of the developers they were working for; but with the growing scale of modern AAA projects, co-development firms are more visible than ever. "There's a growing awareness and desire among studios worldwide to leverage distributed development and external services," says Jake DiGennaro, the chief revenue officer at Virtuos, which, along with Bethesda, co-developed the recently released The Elder Scrolls 4: Oblivion Remastered. "Some reputable studios that have shipped their own IP are even embracing a co-op model, working on original IP development while offering co-development services to others. This approach has long been a tradition in Japan and is now becoming mainstream in North America and Europe." Alejandro Garcia-Tunon, the head of co-dev at Side, adds: "Co-development definitely remains a key part of the game development ecosystem – if anything, more so now than ever before. An increased number of development specialities, as well as scopes that are well beyond what the game industry has seen in the past, have increased costs. "Having a full-time team of experts in each field you'd need to cover just wouldn't make sense, so bringing in an expert is a great way for studios and publishers to tap into that talent without blowing up their budgets." Helping hand The work that co-development companies are being brought in to assist with varies wildly. Sometimes it's individual disciplines (art and animation are the most popular, according to research from the External Development Summit), but for others it is entire segments of a project. "What we are finding is that the balance of the co-development work is changing from coding or feature-focused 'team augmentation' to fully rounded content teams that can take on a whole section of games or even whole games, like we did with the delivery of Lego Horizon Adventures," explains Ashley Liu, the managing director of Keywords Studios' Create arm. Garcia-Tunon adds: "It runs the gamut from individual specialities, like AI, combat, physics, networking, and UI, to taking on entire chunks of a game like a full biome or boss. Sometimes we will cover different aspects even within a single project. Image credit: Konami "When we worked on Silent Hill 2 Remake with Bloober Team, we worked on two boss fights in very different ways. For the fight with Eddie's character, we were creative consultants, advising how we could lift the level from the previous version and make it more of a pivotal moment in the game. With Maria, the final boss fight, Bloober Team already had a vision in mind which we were able to help execute from a technical perspective." Despite co-development companies becoming a bigger part of the industry, the sector isn't immune to the broader economics of the games market. For a variety of reasons, publishers have chosen to be more conservative in funding projects, and fewer games in development means less work for those whose role it is to help out. "Demand grew very strongly over the past decade as games became larger and more complex – developers required additional support," explains Liu. "In the last 18 months, the volume of work reversed due to the overall contraction of the industry, but projects are starting to come back as the industry realises there's been an overcorrection. Players expecting more engaging content will drive some renewed demand." The turbulence in the market, however, does present something of an opportunity for co-development companies. In an uncertain world where publishers are being more risk-averse, partners that provide a great deal of flexibility can be useful. "There are opportunities for co-development, though they may not look like what we've seen in the past," Liu says. "Specifically, as our clients restructure their businesses, there are now more opportunities for partners to fill resource gaps. "There are also opportunities with clients that perhaps would like us to take over their older titles, especially as their best teams move on to new projects." "The headwinds that we face collectively as an industry have reached a breaking point where we can't continue to make games the way we did 20 years ago" Jake DiGennaro, Virtuos Garcia-Tunon adds: "The last couple of years have been an anomaly, no question, but the uncertainty has been really challenging. The co-dev space has historically been growing year over year because the entire industry has been growing – more games equals more people needing help to make games. Our opportunities have come from our ability to quickly adapt to fit client demands." Some co-development companies view themselves as somewhat indispensable. The fact that games now require so many people, so many hours of work, and so many disciplines means that it is tough to make them without external help. "The headwinds that we face collectively as an industry have reached a breaking point where we can't continue to make games the way we did 20 years ago and expect not to run into a wall eventually," DiGennaro believes. "With the rising costs of production, the economics of creating a successful game have become increasingly difficult. "We have to acknowledge the fact that there's something about the way we make games that needs a paradigm shift. While factors like discoverability, delivery, and distribution all play important roles, the process of how we build games is a critical part of the problem," DiGennaro continues. "More than seeing this as an opportunity for co-development companies, I think it's an opportunity to reshape the collective mentality and methodologies of our industry towards increased sustainability and reduced volatility." As always, big games companies are very anxious about their IP and will often go to great lengths to protect it. Before its official reveal, screenshots and marketing materials for The Elder Scrolls 4: Oblivion Remastered were leaked from Virtuos' backend on its Wordpress. "We have very strict processes and information controls at Side that have been honed over decades of work and thousands of games," Garcia-Tunon says. "But our business relies on our reputation – successful projects build great long-term relationships and recommendations from previous partners are worth their weight in gold." Image credit: Keywords Keywords' Liu adds: "Ensuring there are no leaks has always been a challenge in game development, especially for projects that take years to complete with hundreds of people working on it at different times. As with any studio, team composition changes throughout the project lifecycle, making leakages an ongoing area to monitor. "We can leverage experienced global IT and InfoSec departments to protect our client's IP and data. They oversee hundreds of projects, thousands of configurations, and have successfully dealt with a vast number of incidents, allowing us to bring a level of protection surpassing many of our competitors," Liu continues. "In the unlikely event of a data breach, we inform the client first and foremost and together we find a way to resolve the situation. Sometimes, because we are open and honest, the clients trust us more." What comes next Looking to the future, it's clear that co-development isn't going anywhere anytime soon. Modern video games are, by and large, too big and require too much specialist expertise to be made by one company, at least within a reasonable timeframe. Co-dev firms are here to stay. "Fundamentally, I think distributed co-development is how games will be – and should be – built moving forward," DiGennaro says. "It has the potential to make a profound impact in the sustainability of this industry." Liu adds: "Despite the layoffs and project cancellations, this industry will grow. We'll work with partners to release more creative content and better experiences, leveraging new technology to create games and mechanics that have never been seen before. We'll reach new artistic heights that truly engage players." But, as with the rest of the industry, change may be afoot in the near future. New technologies do look like they could have an impact on how games are made. "Workflows are already evolving to include developing technologies like AI," Garcia-Tunon admits. "That said, my outlook is positive – as long as we have humans playing games, we'll need humans to develop and discuss what games should look like."

From Kryptonite bullets to DC comic creator tributes, Superman's latest trailer is full of missable moments.

Menu Home News Hardware Gaming Mobile Finance Deals Reviews How To Wccftech Deals Here Is Another Rare Chance To Grab The PowerColor Reaper Radeon RX 9070 For $679.99 On Amazon And Enjoy High Framerate Gaming At Multiple Resolutions, Including 4K Omar Sohail • May 15, 2025 at 04:04am EDT AMD introduced the Radeon RX 9070 as a mid-range GPU for gamers who wanted to immerse themselves in fluid gameplay sessions at the 1440p and 4K resolutions. Who would have thought that this graphics card punches well above its weight class, because it definitely makes NVIDIA’s more expensive offerings less appealing in value. What is even more appealing is the latest deal on Amazon, where the PowerColor Reaper version of the RX 9070 is available for $679.99. This is only the second time that we have spotted any variant of the RX 9070 available at the aforementioned price, giving you a platinum-level chance to score one of these units and be on your way to enjoying those high-framerate gaming sessions. For those who do not know, the RX 9070 and RX 9070 XT ship with 16GB of GDDR6 VRAM, making them ideal when cranking up those visual settings at the QHD and UHD resolutions. The PowerColor Reaper RX 9070 is also equipped with a triple-fan cooler, and unlike NVIDIA’s top-end GPUs, this one ships with two 8-pin PCIe power connectors. It is recommended that your system is equipped with a 650W power supply unit or higher, and that you have ample space to accommodate the 289mm length of the card. As for the number of display inputs, you get three DisplayPort 2.1 and a single HDMI 2.1, effortlessly allowing for a multi-monitor setup without ever experiencing a performance hitch. What we have always advised potential buyers is to take advantage of a deal as early as possible, and over here, we will reiterate the same message. If you place an order right this moment, you will have to do a little waiting around because the PowerColor Reaper RX 9070’s free delivery will take anywhere between June 1 and June 30 to reach your doorstep. We understand that it is an exceptionally long time, but what you are also getting is an exceptional deal, so if the only trade-off for that is having to muster up a little patience, we would accept it in a heartbeat. Get the PowerColor Reaper AMD Radeon RX 9070 from Amazon - $679.99 See the latest technology deals that Amazon has discounted today See what else Amazon has discounted today Deal of the Day Subscribe to get an everyday digest of the latest technology news in your inbox Follow us on Topics Sections Company Some posts on wccftech.com may contain affiliate links. We are a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for sites to earn advertising fees by advertising and linking to amazon.com © 2025 WCCF TECH INC. 700 - 401 West Georgia Street, Vancouver, BC, Canada

Grapezard Member Nov 16, 2017 8,356 Yup, the legendary dogs beasts cats of Johto. Which is your favorite? You got Entei, the Volcano Pokemon. Suicune, the Aurora Pokemon. and Raikou, the Thunder Pokemon.   mopinks Member Oct 27, 2017 32,684 I'm not crazy about any of em but I'll give it to suicune   AgentStrange Member Oct 25, 2017 4,998 None. I take Zoloft. Entei   DanteSparda504 Member Nov 2, 2017 4,829 Birmingham, AL Suicune is my favorite. I think the only forgettable one is Raikou.   Rustyspider13 Shinra Employee Member Nov 16, 2023 3,584 Suicune, there is a majestic nature about it. Never used any of them in battle though.   RochHoch One Winged Slayer Member May 22, 2018 21,257 Can I cheat and say Raging Bolt What a beautifully silly creature  nsilvias Member Oct 25, 2017 29,958 raikou is where its at. entei is a loser   nsilvias Member Oct 25, 2017 29,958 RochHoch said: Can I cheat and say Raging Bolt What a beautifully silly creature Click to expand... Click to shrink... how do you go from sabre tooth dog to a giraffe   Hero Prinny Member Oct 25, 2017 3,826 DanteSparda504 said: Suicune is my favorite. I think the only forgettable one is Raikou. Click to expand... Click to shrink... Thats cause TPC never did anything with it. Entei got a movie & suicine got a movie and a game. They told poor raikou to go kick rocks   Shuckle Member Mar 30, 2018 598 Suicune is the coolest and was really fun to play in Gen 3 I think. Substitute/Calm Mind/Surf/Rest or Ice Beam if I recall was a really powerful moveset.   RochHoch One Winged Slayer Member May 22, 2018 21,257 nsilvias said: how do you go from sabre tooth dog to a giraffe Click to expand... Click to shrink... Alternate timelines that don't make sense, apparently. Must be from an unga bunga Flinstones universe full of dinosaur Pokemon   TheAggroCraig This guy are sick of the One Winged Slayer Member Nov 6, 2017 7,286 Suicune is the best of the GSC legendaries, but Raging Bolt is by far the best of the paradoxes.   DoctorPlayer MD "This guy are sick" Member Feb 4, 2021 2,728 Brazil All 3 of them are great, but Suicune's my favorite.   Foolish Member Jun 23, 2022 1,507 Okay so Raikou has the most fun name to say. Raikou. Hell yeah. Love it. But I don't like its purple hair that runs down its back. Feels excessive to the design and is way too similar to what Suicune has going on. And the tail feels weirder the more you look at it. Entei has the best design of the lot. Simultaneously comes across as both a lion and also some pure fantasy beast. And those black bands on its legs give the impression that it was once shackled in chains, and has now broken free. So good. I think the purple mane on Suicune clashes way too much with its other colors, and I think its face looks weird. And it cursed my childhood for years because I could not, for the life of me, figure out how its name is supposed to be pronounced. Entei >> Raikou >>>>> Suicune  Faiyaz Member Nov 30, 2017 6,599 Bangladesh Entei is the fucking GOAT   mrmickfran The Fallen Oct 27, 2017 33,133 Gongaga Suicune my beloved   Beren One Winged Slayer Member Oct 25, 2017 4,992 Enter, but that's definitely due to my exposure to the third movie. Suicune is definitely cool and I won't knock anyone for picking that one. Raikou was great competitively but it just couldn't compete. Had no idea what the vibes were.  NetOpWibby Member Nov 15, 2017 75 Carina Nebula SUICUUUUUNE Gimme Pokémon Crystal with Breath of the Wild mechanics and I'm good for a decade.  -Le Monde- Avenger Dec 8, 2017 14,228 My favorite legendary trio. Really hard to choose. Probably Raikou  ForteanMan Member Oct 25, 2017 571 Dongguan, China Think it's Suicune for me largely because of Pokémon Crystal.   StrappingYoungLance Member Oct 27, 2017 7,764 Melbourne, Australia Picking Suicune is for mids.   Bucca Member Oct 25, 2017 5,435 Suicune the GOAT   mnk Member Nov 11, 2017 7,937 Raikou's the only one I like.   Lightsong Member Nov 11, 2022 9,346 Suicune! I also like the dino version:   Dogui Member Oct 28, 2017 10,911 Brazil Suicine is the only not fugly one.   Rocky Road Member Jun 1, 2018 1,006 Suicune is gonna be the most popular due to Crystal, I'm interested in what a straight Entei vs Raikou would be if all the Suicune voters had to vote for one of those two   Buzzkiller_20 Corrupted by Vengeance Member Oct 27, 2017 2,522 I like Entei because of the movie, Suicune design wise, and Raikou because of the bond i had with him in Heartgold. Helped me beat the game and beat Pearl as well when I transferred him over. It's hard to choose which one i like the most, but i think i will give it to Suicune. His design is great and water types are my favorite, so I choose Suicune.   hydruxo ▲ Legend ▲ Member Oct 25, 2017 22,672 They're all about even for me but Suicune has an ethereal vibe to it that puts it over the edge for me   Bishop89 What Are Ya' Selling? Member Oct 25, 2017 42,657 Melbourne, Australia mopinks said: I'm not crazy about any of em but I'll give it to suicune Click to expand... Click to shrink... Exactly this   Stencil Mailing His Own Damn Business Member Oct 30, 2017 13,464 USA I suppose Raikou is a very distant second to Suicune. But Suicune is far and away the coolest and best. Poor Entei.  MimicBox82 Member Jan 9, 2018 5,406 Sweden Suicune is the cutest, so I'll go with that.   HardRojo One Winged Slayer Member Oct 25, 2017 28,818 Peru Raikou, always. RochHoch said: Can I cheat and say Raging Bolt What a beautifully silly creature Click to expand... Click to shrink... Not whatever this monstrosity is. And yeah, someone's gonna claim "well, that's not Raikou".   Shoichi Member Jan 10, 2018 12,459 favorite typing is water pokemon, so Suicune has been one of my favorites since Crystal.   Tsaki Member Feb 12, 2019 419 youngster: Suicune teenager: Raikou adulthood: Entei  Silver9 Saw the truth behind the copied door Member Oct 31, 2017 1,228 Suicune is a majestic pokemon.   Rouk' Member Jan 10, 2018 9,530 To this day, I still don't understand why Raikou is the least popular   Kneefoil Member Oct 25, 2017 4,945 Entei's the coolest, for sure. Suicune's the most iconic these days, though.   onibirdo Member Dec 9, 2020 3,555 I recently watched ProZD's Pokémon ranking videos. He calls Entei "your angry dad as a Pokémon" which is weirdly accurate.