The current wave of generative AI animation often feels like a one-time magic trick. You type in a prompt, a video appears, and if you don’t like the result — maybe the feet are wonky, which is a common problem with generations of AI — your only real option is to try a different prompt. This “black box” approach is exactly what Cartwheel, a new 3D animation startup, is trying to unpack.
Andrew Carr and Jonathan Jarvis, two veterans with roots in OpenAI and Google, respectively, founded the company, working to create a future where AI handles the technical difficulties of animation while leaving the creative soul to the artist.
I spoke with Carr and Jarvis about introducing their company, explaining “taste” and AI, and the technical and creative challenges of animation in 2026.
What makes Cartwheel different
According to the founders, one of the biggest obstacles in this space is that 3D motion data is remarkably scarce compared to the endless seas of text and images available online that AI models are trained on.
“If you look at all the big technology companies, they’ve built their models for written language, sound, image, [and] video because it’s so massive, so finding those patterns is so easy,” Jarvis said. “We knew it was going to be hard, but it turned out to be harder than we thought by maybe a factor of 10 or 100 to find that data.”
Read more: Generative AI in Gaming Is Here, But It’s Facing Reluctance From Gamers — And Developers
While other tech giants focus on producing the last pixel, Cartwheel has spent years mapping how people actually travel. Their models are built to understand the nuances of performance so that a simple 2D video of a person dancing in their backyard can be translated into an accurate, realistic 3D skeleton.
This transition from flat images to 3D objects is what gives animators the control they’ve been missing in the AI era.
Cartwheel has spent years tackling the difficult task of mapping how people actually travel.
Preventing “matching” of AI
Cartwheel executives say they view AI’s “uniformity” as a result of a lack of control. If everyone uses the same generator to produce the video, the results may start to look very similar.
“The output of our system is designed for people to edit. It’s designed for people to touch and manipulate, and we don’t want someone to type something and shove it into a finished animation. That’s not your goal. That’s boring, who’s going to watch that?” Carr said.
“The fact that it’s so easy for people to get into it and sort it out actually completely eliminates the uniformity problem,” he said. “You put it on different characters, you put it in different places, you change the way it looks, you push the action, you pull the action, and so on. [sameness] it turns into nothing.”
Carr and Jarvis said the solution is to provide a “control layer” where the AI’s output is just the beginning. By generating 3D data instead of flat video, the creator can change the lighting, move the camera or adjust the character’s pose after the AI has done its first job — making the technology a sophisticated power tool rather than a replacement for the artist.
Inventor Andrew Carr said that one of his fundamental ideas of science is that motion and movement are fundamental data types.
The future of animation with AI
Besides making animation faster and lowering the barrier to entry, the company is moving toward a concept they call “open storytelling” or “open world building.” In today’s games and social media, the demand for content has reached a scale that hand-made animation will never match.
Cartwheel visualizes characters that are not just programmed with a few movements but powered by motion models that allow them to react and act in real time. It’s less about editing each frame and more about “rehearsing” with a digital actor who understands the intent of the scene.
Ultimately, the goal is to bridge the gap between 2D vision and 3D execution, the inventors say.
“One of the fundamental ideas that we hope holds true for the next three years of Cartwheel is that everyone will be working in 3D whether it’s written in 2D, whether the end result is 2D video,” Carr said.
By focusing on the “layer below the pixels,” Carr and Jarvis say they hope that as animations become more automated, they also become more personal. The machine handles the biomechanics and the files are exported, but the person retains the final word on taste, timing and the heart of the matter.
