# Cartwheel (animation company)

> Source: https://aiwiki.ai/wiki/cartwheel_ai
> Summary: Cartwheel is an American generative AI company that builds 3D character animation tools for games, film and television, advertising, and robotics.
> Updated: 2026-08-07
> Fact-checked: 2026-08-07
> Categories: AI Companies, AI Tools & Products, Computer Vision, Generative AI
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "Cartwheel (animation company)." aiwiki.ai, 7 Aug 2026. https://aiwiki.ai/wiki/cartwheel_ai
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

| Field | Value |
|---|---|
| Type | Private AI company |
| Industry | 3D animation, motion capture, generative AI |
| Founded | 2023 |
| Founders | Jonathan Jarvis (CEO), Andrew Carr (chief scientist) |
| Employees | 15 (February 2026) |
| Flagship model | Comic 4.2 performance capture (August 2026) |
| Other products | Swing, Hermes, Cartographer, Character Generator, Motion Library, Mascot |
| Disclosed funding | US$15.6 million (as of May 2025) |
| Key investors | Craft Ventures, Accel, Khosla Ventures, WndrCo, Tirta Ventures, Runway, Human Ventures |
| Website | getcartwheel.com |

**Cartwheel** is an American [generative AI](https://aiwiki.ai/wiki/generative_ai) company that builds 3D character animation tools for games, film and television, advertising, and robotics.[3] Founded in 2023 by designer Jonathan Jarvis and former [OpenAI](https://aiwiki.ai/wiki/openai) researcher Andrew Carr, the company turns video, text prompts, and a large motion library into rigged, editable 3D character animation that exports into standard production pipelines.[4] Its flagship perception model line, Comic, performs markerless motion capture from ordinary video; the Comic 4.2 update, announced on August 3, 2026, added a foot physics estimation system.[1][2] Cartwheel is unrelated to [Cartwheel Robotics](https://aiwiki.ai/wiki/cartwheel_robotics), the humanoid robot company founded by Scott LaValley.

## Origins

Jarvis and Carr started the company together in 2023.[4] Jarvis was hired into Google's creative team in New York after an animated explainer he made about the 2008 financial crisis went viral; the film was later exhibited at MoMA.[8][21] At Google he worked on design and marketing with the Creative Lab, [Google Brain](https://aiwiki.ai/wiki/google_brain), Search, and Android teams, and he founded and still runs the animation and game studio Universal Patterns.[4][21] Carr studied applied mathematics and computer science at Brigham Young University, interned at Google Brain and Lyft's autonomous vehicle division, and worked at OpenAI in 2021 as a fellow on the code generation team, where he built evaluation datasets for the [Codex](https://aiwiki.ai/wiki/openai_codex) models behind [GitHub Copilot](https://aiwiki.ai/wiki/github_copilot) and is a co-author of the Codex paper "Evaluating Large Language Models Trained on Code".[10][11] Before Cartwheel he was a senior applied research scientist at the synthetic data company Gretel.ai.[10]

Carr described the founding motivation to Forbes: "There's this core frustration animators face with today's software. You need huge teams and weeks of effort to do things that should be fast. We wanted to solve that."[4] TechCrunch profiled the product in June 2024, when it generated basic humanoid animation from a text box in a minute or two; Jarvis framed the problem as a "big 'blank screen' problem, where there's a million buttons and options".[6]

The company came out of stealth and released its first public tools in May 2025, after a year-long early access program that Forbes reported drew more than 8,000 beta users, including creatives from DreamWorks, Roblox, Duolingo, and Take-Two.[4][7][9] Pixar veterans joined at the same time: animation director Catherine Hicks as head of animation innovation, and Neil Helm as head of interactive animation.[4] GamesBeat describes Helm as a 16-year Pixar veteran who animated on Up, Turning Red, Coco, and Lightyear and served as crowds animation supervisor on Toy Story 4 and Inside Out 2.[8] As of February 2026 the company had 15 employees.[8]

## Products

Cartwheel sells a browser-based workspace, Cartwheel Canvas, plus an API for studios that want to generate motion inside their own tools and pipelines.[3][8] The founders told GamesBeat their tools can make some animation tasks as much as 100 times faster than manual workflows, a vendor claim that has not been independently benchmarked.[8]

| Product | Function |
|---|---|
| Comic | Performance capture model: turns video into character animation without suits, markers, or studio hardware[2] |
| Swing | Text-to-motion generation, producing multiple takes from a prompt in seconds[12] |
| Hermes | Motion editing model: repose the character directly and the model fills in surrounding motion, which Cartwheel compares to inpainting for movement[3][8] |
| Cartographer | Universal retargeting: maps motion across rigs with mismatched proportions, from human bipeds to quadrupeds or stylized characters[13] |
| Character Generator | Generates rigged, animation-ready 3D characters from text descriptions or reference images[15] |
| Motion Library | Curated library of more than 27,000 motions delivered as editable data in standard 3D formats[14] |
| Motion Editor | Path and timing editing over existing motion data without discarding the original[20] |
| Mascot | Turnkey character content for social media and brand marketing[16] |
| API | Programmatic generation for batch jobs such as locomotion sets, gesture libraries, and crowd motion[3] |

The platform exports rigged motion in standard formats, including FBX with presets for Blender, Maya, Unreal Engine, and Roblox, plus GLB, BVH, USDZ, Alembic, Collada, and MP4; Maya scene files are reserved for paid plans.[3] Access runs through free and paid subscription tiers with monthly generation credits, alongside team and API plans.[3] The tooling is closely integrated with Unreal Engine.[8]

## The Comic model line

Comic is Cartwheel's markerless video motion capture family, which Jarvis calls the company's "flagship perception model".[1][2] Instead of a mocap stage with suits and camera arrays, it extracts skeletal motion from footage shot on any camera, including phones, and retargets it onto arbitrary characters.[2][8]

Company case studies describe Comic 3 in production use during 2025 and 2026. The animation studio Psyop used a hybrid workflow built on Comic 3 to repair unstable arm and body motion in an existing motion capture shoot without reshooting the performance.[18] A team led by Academy Award winning director Mark Andrews (Brave) used Comic 3 with Unreal Engine on Dogstar, an episodic social media animation project: the team acted out scenes on phone cameras and retargeted the footage directly onto the show's characters, compressing production cycles from weeks to days.[17]

Comic 4, the current generation, added multi-person capture from a single wide shot, camera intrinsics and extrinsics estimation so motion lands in correct spatial relationship to the scene, facial performance and finger-level hand capture from one video source, and relative scale capture.[2]

On August 3, 2026, Jarvis announced Comic 4.2, focused on ground contact: "Feet on the ground. It's always been one of the hardest problems in capturing human motion." He said the update centers on "an advanced foot physics estimation system, properly estimating force, friction, toe roll, and elevation".[1] Foot sliding and ground contact errors are a long-standing weakness of monocular [pose estimation](https://aiwiki.ai/wiki/pose_estimation), the [computer vision](https://aiwiki.ai/wiki/computer_vision) task of reconstructing 3D joint positions from 2D video, which offers no direct force measurement.

## Technology and training data

Craft Ventures, the company's lead investor, says Cartwheel is not built on a large language model but on a "purpose-built generative engine" it describes as Large Motion Models, with a proprietary motion data structure designed by Carr.[5] Carr told TechCrunch the underlying model is original and that motion is represented as a matrix of joint positions over time.[6]

Both founders point to licensed training data. Carr said in 2024, "We have multiple data sources, all ethically sourced, and our own labelers labeling these motions," and Forbes reported that the character generator is trained on licensed 3D assets.[4][6] Asked by GamesBeat whether actors might object to captured motion being edited into performances they never gave, Jarvis said the company pays actors for the use of their likeness and added: "We deal in motion. We're not re-creating the likenesses of other people or characters or IP. We have guardrails to prevent the generation of a character for something like that."[8]

## Funding

| Round | Announced | Amount | Lead | Other participants |
|---|---|---|---|---|
| Seed | June 2024 | US$5.6 million | [Accel](https://aiwiki.ai/wiki/accel) | [Khosla Ventures](https://aiwiki.ai/wiki/khosla_ventures), Human Ventures, Heretic VC, MVP Ventures, Correlation Ventures, Pelion VC, angels[6] |
| Seed (second) | May 2025 | US$10 million | Craft Ventures | WndrCo, Tirta Ventures, Runway, Accel, Khosla Ventures, Human Ventures[4][7] |

The May 2025 round, led by Craft Ventures with participation from Jeffrey Katzenberg's WndrCo, Ben Feder's Tirta Ventures, and the AI video company [Runway](https://aiwiki.ai/wiki/runway), brought total disclosed funding to $15.6 million.[4][5][7] GamesBeat reported the total as $16.6 million in February 2026.[8]

## Robotics and motion data

Cartwheel describes itself as "a generative motion platform for games, film & TV, advertising, and robotics", and its hiring materials talk about building motion tools for "games, film, robotics, and embodied systems".[3][19] The company argues on its site that [humanoid robots](https://aiwiki.ai/wiki/humanoid_robot) will need large volumes of human motion data, particularly for non-verbal communication, and that motion captured from video can be retargeted onto robot embodiments the same way it is retargeted onto game characters.[3] Jarvis made the same case to GamesBeat: capture, generation, and editing feed a universal retargeting pipeline that "is also quite helpful in the field of robotics, where they need a lot of nonverbal communication data".[8] The site markets pre-built motion datasets and custom data programs alongside the animation tools.[3]

## Competition

Forbes placed Cartwheel in a field that includes Anything World, which auto-rigs and animates 3D assets for game engines, and Wonder Dynamics, the Autodesk-owned system that inserts CG characters into live-action footage.[4] In markerless video motion capture specifically, Cartwheel competes with dedicated capture vendors; a customer testimonial the company publishes contrasts it with Move AI on cost.[3] Unlike video generators such as Runway, its output is rigged, editable 3D motion data rather than rendered video, aimed at animators who finish the work in tools like Maya, Blender, and Unreal Engine.[4][6]

## See also

- [Cartwheel Robotics](https://aiwiki.ai/wiki/cartwheel_robotics)
- [Pose estimation](https://aiwiki.ai/wiki/pose_estimation)
- [Runway](https://aiwiki.ai/wiki/runway)
- [Physical AI](https://aiwiki.ai/wiki/physical_ai)
- [Tripo](https://aiwiki.ai/wiki/tripo)
- [World Labs](https://aiwiki.ai/wiki/world_labs)

## References

1. Jonathan Jarvis (@JonathanJarvis), X post announcing Comic 4.2, August 3, 2026. https://x.com/JonathanJarvis/status/2084391878397432062
2. "Comic 4", Cartwheel product page. https://getcartwheel.com/product/comic-ai-video-motion-capture
3. Cartwheel official site. https://getcartwheel.com
4. Charlie Fink, "AI Startup Cartwheel, Led By OpenAI And Google Veterans, Raises $10 Million To Simplify 3D Animation", Forbes, May 15, 2025. https://www.forbes.com/sites/charliefink/2025/05/15/ai-startup-cartwheel-led-by-openai-and-google-veterans-raises-10-million-to-simplify-3d-animation/
5. "Why We Invested in Cartwheel", Craft Ventures, May 2025. https://medium.com/craft-ventures/why-we-invested-in-cartwheel-1a4e99824673
6. Devin Coldewey, "Cartwheel generates 3D animations from scratch to power up creators", TechCrunch, June 5, 2024. https://techcrunch.com/2024/06/05/cartwheel-generates-3d-animations-from-scratch-to-power-up-creators/
7. "AI Startup Cartwheel Raises $10 Million; Releases Its First Character Animation Tools (EXCLUSIVE)", Variety, May 15, 2025. https://variety.com/2025/artisans/news/ai-cartwheel-raises-10-million-character-animation-1236395878/
8. Dean Takahashi, "Cartwheel claims AI can make 3D animation tasks as much as 100x faster", GamesBeat, February 25, 2026. https://gamesbeat.com/cartwheels-ai-can-make-3d-animation-tasks-as-much-as-100x-faster/
9. "AI Animation Firm Cartwheel Emerges From Beta; Pixar Alums In Exec Ranks, Jeffrey Katzenberg's WndrCo Among Backers", Deadline, May 2025. https://deadline.com/2025/05/ai-animation-firm-cartwheel-emerges-from-beta-pixar-jeffrey-katzenberg-1236404797/
10. Andrew Carr, curriculum vitae. https://andrewnc.github.io/CV.pdf
11. Mark Chen et al., "Evaluating Large Language Models Trained on Code", arXiv:2107.03374, July 2021. https://arxiv.org/abs/2107.03374
12. "Swing", Cartwheel product page. https://getcartwheel.com/product/swing-motion-generation
13. "Cartographer", Cartwheel product page. https://getcartwheel.com/product/cartographer
14. "Motion Library", Cartwheel product page. https://getcartwheel.com/product/motion-library
15. "Character Generator", Cartwheel product page. https://getcartwheel.com/product/character-generation
16. "Mascot", Cartwheel product page. https://getcartwheel.com/product/mascot
17. "Dogstar", Cartwheel case study. https://getcartwheel.com/blog/dogstar
18. "Psyop", Cartwheel case study. https://getcartwheel.com/blog/psyop
19. "AI Scientist (Computer Vision & 3D Understanding)", Cartwheel job listing. https://getcartwheel.com/openings-ai-scientist-computer-vision
20. "Motion Editor", Cartwheel product page. https://getcartwheel.com/product/motion-editor
21. Jonathan Jarvis, personal site. https://jonathanjarvis.com/