A field guide to the video diffusion zoo

What it does This is an awesome-list that tracks the rapidly expanding universe of diffusion-based video research. It collects open-source toolboxes, foundation models, evaluation benchmarks, and commercial products, then sorts them into granular categories like controllable generation, motion customization, long-video generation, and even game generation. If a paper uses diffusion to synthesize, edit, or restore video, it is probably listed here.

The interesting bit The taxonomy is the real product. The maintainers slice the field into surprisingly specific niches—separate sections for talking heads, virtual try-on, 4D generation, and AI safety—turning a chaotic literature stream into a browsable directory. It is essentially a human-curated search index for a subfield that moves faster than arXiv filters.

Key highlights

• Covers the full pipeline: foundation models (e.g., Cosmos, HunyuanVideo, Wan-Video), inference frameworks (FastVideo, LightX2V), and evaluation metrics (LikePhys, Stable Cinemetrics).
• Sections extend well beyond pure generation into video editing, audio synthesis, talking heads, 3D/4D, game generation, and healthcare applications.
• Includes both heavyweight industrial releases (NVIDIA, Meta, Tencent, Stability AI) and smaller academic toolboxes.
• Explicitly tracks commercial products alongside open-source code, which is useful for benchmarking what is actually shipping versus what is merely published.

Verdict Researchers and engineers building video-generation pipelines should bookmark this. If you are looking for a tutorial or a unified framework, this is just a well-organized index—not the destination.