AnimateDiff

Released in mid-2023 by an academic team led by Yuwei Guo, AnimateDiff showed that one separately trained motion module could be slotted into almost any Stable Diffusion 1.5 checkpoint, so community fine-tunes from sites like Civitai could suddenly produce short animated clips with no per-model training. Later versions added MotionLoRA camera moves, SparseCtrl sketch and RGB conditioning, and a beta SDXL module, and the idea was copied widely across open video tooling. The official repository stopped receiving commits in July 2024, and community pull requests have piled up unmerged since. It still works: the motion weights remain on Hugging Face under Apache-2.0, Diffusers ships AnimateDiff pipelines, and the ComfyUI-AnimateDiff-Evolved node pack was still being updated in 2026, which is where most people run it now. Expect clips of a few seconds at SD 1.5 resolutions with some flicker. For new projects, dedicated open video models such as Wan 2.2 or LTX-Video give longer and more coherent results, while AnimateDiff remains handy for animating one specific SD 1.5 style.