Making Flow Interview with director Gints ZilbalodisJanuary 22nd, 2025User StoriesFrancesco Siddi html PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN" "http://www.w3.org/TR/REC-html40/loose.dtd"Flow, the animated feature film following the mystical journey of a dark grey cat and his companions, is the manifestation of Blenders mission, where a small, independent team with a limited budget is able to create a story that moves audiences worldwide, and achieve recognition with over 60 awards, including a Golden Globe for Best Animation and two Oscar nominations.In this interview, Gints Zilbalodis, writer and director (and more!) of the film, shares how Blender was instrumental in the creation of the film.Gints: Ive done animation, all kinds of animation. I started doing hand drawn, 2D, digital animation. But after making a few shorts, I realized that Im not good at drawing, and I switched to 3D because I could model things, and move the camera. And so at first, I used Maya, which was taught at our school at that time.After finishing my first feature Away, I decided to switch to Blender in 2019, mainly because of EEVEE. I started using the 2.8 beta or even alpha release. It took a while to learn some of the stuff, but it was actually pretty straightforward. Many of the animators in Flow took less than a week to switch to Blender.EEVEE was interesting to me because, even my first feature Away, it was all playblasted, which is not like proper rendering, rather its like previews.I was excited to find that workflow in Blender, but in a more advanced way that gave me greater control. Speed is really important to menot just in rendering but also in working with files, setting up lighting, and creating the overall look. I like to work on multiple aspects at the same time; for example, when setting up the camera, I also need lights in place because lighting influences camera placement and how the scene looks. Thats why EEVEE was so appealing to me.I briefly experimented with some game engines, but at least back then, it was really difficult to figure out a workflow for making films in them.And Blender was ideal: it had all the tools I needed.The entire project took about five and a half years. In the first year, I was writing the script, learning Blender, and looking for funding as Dream Well Studio. That was in 2019.In 2020, we secured some funding, and I moved into a co-working studio space with other artists and developers who were using Blender. Thats where I connected with Mrti Uptis and Konstantns Vievskis.Mrti was one of the first people I approachednot specifically for water simulation, but just to see how he could contribute. However, it quickly became clear that he had a deep expertise in water, unlike anyone else.We were fortunate that, in the early stages, it was just me, so the pandemic didnt affect us much. By the time we moved into full production in 2023, things had stabilized.I created a short pilot for Flow about a minute and a half long where I went through the entire workflow. It was technically basic, but it was useful to test the process. That led to our first teaser, which I never showed publicly. Later, we made another, entirely new teaser, which we used for pitching.In 2021, we started hiring concept artists and building the team. We brought in riggers and developers to create custom scripts that helped streamline the workflow while I was working on the animatic.The Latvian studio was relatively small, it all fit in one room. In total, we had around 15 to 20 people, but at any given time, there were usually only three to five people working, since different teams handled pre-production and post-production.We had a set-dressing team. I would design the initial scene in previz, and they would refine it by adding more plants, props, and environmental details. Concept artists sketched out buildings and figured out their construction, incorporating storytelling elements into the environments.Other team members focused on developing tools. Water was a huge part of the film, but only two people handled all the water effects. Mrti had already been researching water simulations and posting his findings on YouTube, but he hadnt yet put everything together. He eventually developed a Blender add-on for water effects.Meanwhile, Konstantns handled smaller simulations, such as splashes. He also researched techniques for stylized fur and feathers, working on shaders. In addition to that, he did rigging, and character modeling along with other team members.In 2022, Belgian and French co-producers Take Five and Sacrableu Productions joined the project to work on sound, character animation and additional aspects of the film. Expanding the team with experienced character and pipeline TDs, as well as animators working in a well-structured process, was essential to handle the complexity required by the film. This was a truly international coproduction.The film premiered at the Cannes Film Festival in the Un Certain Regard selection in 2024.How did you learn Blender?I learned a lot online, but it was great to have someone with more experience next to me (Konstantns). He did a lot of rigging and was much more technical than me, so I could ask him for advice. Sometimes, I needed something specific in the animatic, like the deer moving in a spiral, and he would write a script to automate it. This was before Geometry Nodes.I cant write scripts myself, so having someone in the studio to help was invaluable. But learning never really stops. I still feel like theres so much I dont know about Blender or anything else. And with these long projects, you sometimes forget things you learned five years ago.Flow was made entirely with Blender and rendered with EEVEE. Each frame took from about 0.5 10 seconds to render in 4k. We didnt use a renderfarm. The final render was done on my PC. There was no compositing, all the colors were tweaked and adjusted using shaders.How does the previz process work?When creating the previz or animatic, I just try to get things done as quickly as possible. This approach helps me explore ideas efficiently. Im not great at drawing, so previz works better for me. Its faster, and I like to move the camera a lot. Sometimes, I roughly sketch out a building, but its often very basic.I then hand these files over to a concept artist. Many environmental concept artists use Blender as well, so they can import my files. While they usually rebuild everything from scratch, my files at least provide the correct proportions. Sometimes, they paint over my models, but in other cases, they design everything directly in 3D.When they send the files back, I ask them to leave assets in place rather than moving them to the center of the scene. That way, I can easily import everything back, and it aligns perfectly.The animation teams in France and Belgium brought a great deal of organization to the process. They developed further tools and rigs to deliver character animation, they had to optimize the scenes, removing everything except the assets the characters interacted with and cleaning them up thoroughly. However, I didnt use these optimized assets directly, I would import their animations back into my heavier scenes.For lighting, it was just me. We had other people handling different tasks, but I was solely responsible for lighting. This setup made things easier.Since I handled a lot of tasks myself, it was simpler to work with large files where everything was imported. In each file, I made extensive adjustments to assets. For example, when setting up lighting, I tweaked materials for the assets in each shot, making them slightly lighter or darker to get the right look. I know this could be done with library overrides, but I was also working across different computers: my desktop PC and my MacBook.Switching between operating systems sometimes caused issues with linked assets, even when using relative file paths. To avoid breaking links, I found it easier to keep everything within the file itself. Some of the smaller scenes were around 300 MB compressed, while a few of the largest ones reached nearly 2 GB compressed.Maybe I could have figured out a better way to link assets, but during production, speed was the priority. The production timeline required me to move fast, so I opted for the most efficient workflow rather than experimenting with alternatives.Learn more about the animation of Flow in this Blender Conference presentation by Animation Supervisor Lo Silly-Plissier.A glimpse into the water surface system used in Flow.As an early adopter of Blender 2.8, did you upgrade as new releases became available?I started with Blender 2.8 alpha while it was still in development, and I was constantly updating things. I think when the team joined, we were using 2.9 or maybe 3.0.With each major version, we decided to update since there were only a few of us at the time, and we werent sharing files. That made it safer because everyone was working on their own files independently, without links. The last version we used was 3.6. EEVEE definitely improved over time, but it wasnt just EEVEE. Geometry Nodes and other features made upgrading worthwhile.Of course, before each update, we ran a lot of tests, opening different files to check for issues. Some things did break, but overall, our workflow remained stable.Early on, when the team was small, updating wasnt a big deal. But once all the animators started in 2023, they worked in 3.3 and stuck with it throughout production. After they finished animation and I moved on to lighting, I imported everything into 3.6, which wasnt a problem.Which add-ons were part of your workflow?We used a few. One of them was GeoScatter, a popular scattering add-on for distributing plants and other environmental elements. We also used Animation Layers, not for character animation, but for the camera, specifically to create handheld, shaky camera movements.I created separate layers for a standstill shot, for walking in place, and for running. This allowed me to mix and adjust them as needed. I believe some add-ons have been developed since then specifically for this kind of workflow. To generate camera motion, I also tested VirtuCamera. I experimented with recording live camera movements by walking around, but I found it too imprecise. Instead, I preferred keyframing and layering different types of motion.For fluid simulations, we sometimes combined different techniques, starting with large-scale waves using Cell Fluids and then adding details with FLIP Fluids.Other tools we used included Bagapie Vegetation Generator, Bagapie Rain Generator, and Copy Global Transform.What I love is how fast the files open. It might seem like a small thing, but it actually saves a lot of time and frustration.EEVEE is great. Also, I love how customizable everything is. I created a lot of custom keyboard shortcuts, which worked really well when I was working alone. However, once we started working in the studio, it caused some issues, especially when I had to demonstrate something on someone elses computer. But we figured it out.I also love the amount of resources available online. There are so many tutorials and tools, and I can quickly find answers to almost anything.What could improve in Blender for indie filmmakers?Well, there were some challenges with using Blender, but we solved them.Sometimes things werent clear at first, but once you actually put your mind into it, you can figure it out. Thats often the case with Blender: you encounter obstacles, but with enough effort, you find a way through.What Id love to see and I think its already happening is more focus on NPR (non-photorealistic rendering) workflows, which is great. Further improvements to interactive and real-time rendering would also be a huge benefit.I havent worked much in Blender over the past six months, but Im already working on my next project, and I plan to use Blender for it.Final thoughts?Ive never worked in a big studio, so I dont really know exactly how they operate. But I think that if youre working on a smaller indie-scale project, you shouldnt try to copy what big studios do. Instead, you should develop a workflow that best suits you and your smaller team.In our case, we didnt rely heavily on concept art. We modeled the characters directly in 3D and found ways to skip certain steps. Many of us wore multiple hats, figuring out how to streamline tasks rather than having separate departments for everything.For me, its also easier to handle the camera and lighting simultaneously rather than treating them as separate stages. Having a smaller team made the process more flexible and efficient.When developing my first feature, I structured the story around elements that were relatively easy to animate. I avoided large crowds and complex effects because, in the end, most viewers dont think about how difficult something was to create. I think its valuable for filmmakers to collaborate with tool developers early on to understand which things are challenging and which are easy. This can actually spark creative ideas rather than feeling like a limitation.Storytelling offers infinite possibilities, but sometimes constraints can be beneficial. For example, deciding to use only four characters and a handful of locations can lead to stronger creative choices. Some of my favorite films take this approach. They dont need an epic scope to be powerful.That said, I think a certain level of naivety is necessary when starting a project. If I had known how difficult it would be, I might never have started. But because I didnt fully grasp the challenges ahead, I just dove in and figured things out along the way.