ToonCrafter is a generative cartoon interpolation model that creates smooth in-between animation from two cartoon keyframes. Presented at SIGGRAPH Asia 2024, it tackles the classic in-betweening problem — drawing the frames between two poses — with a video diffusion model adapted to the cartoon domain. The project has about 6,000 stars on GitHub and is released under the Apache 2.0 license.
Traditional interpolation relies on optical flow, which breaks down on cartoons: characters deform non-rigidly, lines are sparse, and large movements create occlusions that flow cannot explain. ToonCrafter instead repurposes a pretrained video diffusion prior, rectified for cartoon data, so the model hallucinates plausible intermediate frames that respect the drawing style of the endpoints. It can also be driven with sparse sketch guidance, letting an artist drop a rough intermediate sketch to steer the motion, and it supports related tasks such as sketch-based colorization workflows.
The released repository includes the pretrained model, inference scripts and a demo path so artists can try interpolation without training anything. Practical expectations matter: generation runs at the resolutions and clip lengths typical of research models, and results are strongest on line-art styles similar to the training data. Even so, as a free, local, Apache-licensed model it occupies a niche that commercial animation tools have barely touched — automatic in-betweening that a studio can run and fine-tune in-house.
Free. ToonCrafter is open source under the Apache 2.0 license and runs on your own GPU.
Animators, animation-tool builders and generative-video researchers exploring AI-assisted in-betweening for 2D cartoon production.