The Auteur, The Cat, and The Flood of GPUs
How one guy and some GPUs invented a new waterfall method to make the movie that beat Disney, Pixar, and Dreamworks.
Last night I watched a movie called Flow. It recently won both the Golden Globe and Academy Awards for best animated movie, beating Moana 2, Inside Out 2, The Wild Robot, and others. It has earned $19M worldwide so far (Mar 12 2025), compared to $1B for Moana 2, $1.7B for Inside Out 2, and $0.3B for The Wild Robot. But in its native Latvia, Flow is the highest-grossing film of all time, selling more tickets than Titanic.
I liked it a lot, though I liked The Wild Robot even more (I haven't seen any of the others, no spoilers ahead). For me, Flow’s great achievement is that it has two different characters who felt simultaneously novel and mythic, which seems very hard to do!
The first of these innovative characters initially presents as a standard Jesus / Galadriel / Maximus Meridius sort of guy. But whereas this archetype is usually embroiled in a giant societal-scale conflict, Flow focuses its time with this character entirely on day-to-day life — the mundane consequences of irreversibly sacrificed privilege. And he’s not even the main character!
The second mythic sort of character felt absolutely brand new to me, I can think of no parallel from other fiction, and I still feel such a deep resonance with it that I am convinced there has been an “unnamed archetype” lurking below the cultural water waiting for its true name to arrive in this movie. Sort of a very-low-frequency version of a Greek god.
Everything the movie does, it does in two scales at once. An epic larger-than-life scale, and a simultaneous smaller-than-life ship-in-a-bottle. The exaggerated scale of each accentuates the other, making an experience that’s at once more mundane than Seinfeld and (almost but not-quite) more galactic than Interstellar.
We opened with earnings, not let’s talk cost. In terms of production budget, Moana 2 was $150M, Inside Out 2 was $200M, and The Wild Robot was $78M. Flow cost only $4M. Despite this ~50x cost advantage, Flow still has a lower ROI than its fellow award nominees (for now). Hopefully its growing prestige will help it to reach a much bigger audience, although the awful SEO properties of its name are going to cost it dearly. I loved the movie but I don’t understand how the name relates to it at all. “The Cat and The Flood,” boom, you could even buy the .com for $15.
The reason I’m writing about it is because the filmmaking procedure for this movie is (to me) totally novel, and perhaps the best example of a well-executed “waterfall” project I have ever heard of. It also gives some hints at what generative AI is going to let individuals and small teams make. Just as important, generative AI played no role in the film. Just plain-old non-AI GPUs can enable a fanatic to make a feature-length animated movie for a few million dollars that beats Disney and Pixar along some important axes.
Here’s the story: a 17-year-old Latvian guy named Gints Zilbalodis made a 7-minute animated movie called Aqua and put it on his YouTube. It’s a cat, there’s a flood, the cat is afraid of water, a boat drifts by at the last moment and the cat climbs on, the cat has a mystical relationship with a bird that teaches him how to fish, the flood recedes, then a new flood comes. Every shot is the most dramatic shot that 17-year-old Zilbalodis could think of, sort of spamming the dramatic lighting + vignette button.
When he turned 24, Zilbalodis released a feature-length movie called “Away”, which is currently streaming free on Amazon Prime. He wrote and animated the entire thing himself, even wrote the music. Having shown that he could make at least some kind of movie literally by himself (6.7 out of 10 on IMDB with 1,800 voters), he was able to secure a small amount of funding to expand his old “Aqua” short into the feature length movie that just beat Disney, Pixar, and Dreamworks. This was the first time Zilbalodis had ever worked with anybody else in a professional setting! Four million dollars is ~50x too little to make a feature-length animation, but also 100x too much to give to a never-had-a-job 24 year-old! How did he get it? In his own words, “maybe they thought if there’s a big catastrophe and we run out of budget, I might still finish the film myself in some way.”
There was no catastrophe, but it wasn’t a one-off miracle either. Zilbalodis is going to be able to do this over and over, and other people are too. The first thing to know is that his first feature-length movie, Away, was rendered in realtime! He basically just screencapped the preview window in Maya! And it’s beautiful!
For his second feature film, Flow, he once again sort of did the entire movie himself, but this time he only made rough environments and animatics. Since he was no longer responsible for character-level animation, Zilbalodis “didn’t limit myself this time, setting it around things that are easy to animate.” The difference between a solo auteur and $4M of skilled human labor and taste is this:
I don’t mean to diminish the human contributions of the ~fifty people who earned the $4 million. Zilbalodis certainly doesn’t! The film was cowritten by Matiss Kaza, and the character animation plays a huge role in the emotional impact of the film. But listen to how they did it, in Zilbalodis’ words (he was 24 when this project started, but now he is 30):
(from 3DVF interview)… the team was quite young. Again, we didn't have a huge budget so we had to find creative ways. So a lot of the team, for them it was one of the first projects they've done. And I think that helped also the film because the team was very passionate. For me it was also less intimidating. We were all kind of at a similar age. I didn't feel kind of afraid of working with these established professionals.
…I did the animatic myself and we didn't do any storyboards. …Basically when I'm doing the animatic I create a rough version of the environment. Very Simple. And I explore the environment with this virtual camera and I look for ideas and find shots. I discover things. I know that some artists and some filmmakers, they can really visualize the scenes in their head. But I don't have this ability. I have to discover them. And once I have figured out the camera, we only do the concept art later. And we design the things that are really visible in the camera. So we don't spend any resources on things that will not be in the frame. And we only put detail on the things that are in the foreground. We don't design something that's far away. Everything is designed from the point of view of the camera.
It’s almost the opposite of George Lucas and Star Wars. Instead of a world looking for an excuse to tell a story, it’s a story conjuring up just enough world to get by.
But lots of studios (maybe all other feature animation studios?) use storyboards, which do exactly that - isolate the work of the story from the work of everything else. What’s unique here is that Zilbalodis binds and entangles the story with the camera motion and set design, very early on in the process. Relative to a team batting around a story board, this is a tightly constrained process. But from the perspective of editing and camerawork, it is incredibly freeing. And it shows!
(from Next Best Picture)… And sometimes I might start a shot and it will be just five seconds long, but sometimes I have these ideas and … almost like a stream of conscious … keeps growing and ends up being five minutes long and the camera is following these characters and that's really difficult to draw as well where the the perspective keep changing so much and … it's almost like this dance between the camera and the characters, and having these longer takes it really helps that immersive aspect where you're feeling very close to the characters. You're not observing them from a distance, and you can … create this subjective point of view.
These “stream of consciousness” shots are the sort of thing that a storyboard would kill. The approach is not without its downsides - there are definitely aspects that feel amateurish. The throughline of the movie is incredibly cohesive, but there are some “side groups” who don’t make much sense to me. It might be that I just don’t get it, but I also think that at Disney they’d have a whole team designing the world of each side group, and they’ll sacrifice a little cohesion in the primary narrative arc to preserve integrity to all the little worlds it is traveling through. Sure, universe integrity is great for seeding sequels and theme-park spinoffs, but it’s also good for the movie itself too!
Zilbalodis is a special talent, and his movie is worth watching for its own sake. But what I really want to think a little about is - if he used Sora or Runway instead of human animators, what would he have made?
I think some big films, because the team is so big, they might lose some of those imperfections, some of those rough edges which I like. You can feel that they're handmade. I wanted to have this balance where things are intended, everything is intentional, but also there are some moments where it's pure emotion. It's something that you can't really explain, you just have to feel it and experience it.
AI videos certainly do not feel handmade right now. But just like the big AI labs, Zilbalodis trained his animators on YouTube data
So we looked at references for everything, for every small moment. When the cat is turning its head a little bit, we find some references on YouTube.
And just like the big AI labs, you can tell Zilbalodis was also heavily trained by video games. The opening sequence feels a bit like upmastered Tarzan 64, and many reviews of the movie note the videogame vibe. Famed video game designer Hideo Kojima called Flow his “No. 1 animation film of 2024”, and heaped even more praise on Zilbalodis’ earlier pure-solo work Away:
Removing all the dialogues and connecting the movement, color and sound with camera work are the fruit of imagination. Unique sense and mysterious feeling/sensation. The boy's journey and lonely adventure. The beauty and desolation created by eerie silence. These feelings of nostalgia, the unknown "away" feeling, I find it irresistible. This is another "DEATH STRANDING".
As a computer programmer, it is clear that everything about my industry is going to be different because of generative AI. Sure AI can help with our existing workflows, but the real earthquake is that you don’t need those workflows anymore. You can follow totally novel processes to achieve totally novel results. I can see it in my field, and I’m primed to see it in everyone else’s field too. So when I heard about this one guy making whole animated movies himself, it’s sort of grounding to know that, yes, gobs of compute is changing everything. But no, it’s not just AI, and the same novel forces that are shaping video AI (YouTube data and synthetic videos from game engines) are shaping the next generation of human filmmakers too.
If you like this post, you might also like Disneyland on VHS
Who knows what’s coming next for programmers or animators. But what next for Zilbalodis?
We just finished "Flow" literally days before the premiere in Cannes and I'd never seen it with an audience and that was a big relief to see it in Cannes and when people were laughing at the right places and responding correctly. So it doesn't just means that they understand the film, but it's a more personal feeling that they understand me because it's like a personal story about my own experience and it's a very good feeling to know that you're understood.
… [For my next project] The technique will be a little bit different and I think it's something that's never been done in animation which I will not reveal yet but it's a very ambitious and technically complicated technique.
Places to learn more:
Print interviews - Fast Company, Hollywood Reporter
YouTube interviews - Janus Films, 3DVF, Next Best Picture
Curious to know who the second mythic character is.