FastVideo
Project record
Maintained by Hao AI Lab, UC San Diego156
FastVideo is a post-training and inference framework for accelerated video generation, maintained by the Hao AI Lab at UC San Diego. Its README lists full and LoRA fine-tuning of open video diffusion transformers, Distribution Matching Distillation (DMD2), Video Sparse Attention and sparse distillation, causal distillation through Self-Forcing, and sequence-parallel distributed training and inference.15
- Repository: Repository (hao-ai-lab/FastVideo) (external site: github.com)
- Documentation: Documentation (external site: hao-ai-lab.github.io)
- License: LICENSE (Apache 2.0) (external site: github.com)
- Release notes: Releases (external site: github.com)
Availability and license
Overall availability
Source code is public on GitHub; the README documents installing the package with uv or pip.1
Availability is separate from permission: read the license before using or redistributing.
Component reuse rights
- weights
- Unknown — no complete fact-level rights review
- code
- Reviewed qualifying license recorded — check scope and conditions
- data
- Unknown — no complete fact-level rights review
- documentation
- Unknown — no complete fact-level rights review
No complete system-rights review is recorded for this release.
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| Source codeIs the source code publicly readable? | Public | Public GitHub repository under the Apache License 2.0.12 |
| DocumentationIs user documentation published? | Public | The documentation site covers installation, an inference quick start, training workflows, optimizations, and a support matrix; a cookbook gives model-specific recipes.31 |
| InstallationAre installation instructions or packages publicly available? | Public | The README recommends creating a uv environment and installing the fastvideo package with a CUDA 12 or CUDA 13 PyTorch backend. Apple Silicon users follow an MLX install guide, and NVIDIA DGX Spark (ARM64) requires an editable install from source.1 |
| Supported platformsAre supported operating systems or hardware documented? | Public | The README lists H100, A100, and RTX 4090 GPUs and Linux, Windows, and macOS, plus MLX on Apple Silicon, and links a support matrix of models, hardware assumptions, and optimization compatibility.13 |
| Release statusAre versioned releases published? | Public | Versioned GitHub releases are published; the most recent listed is v0.2.0, released June 4, 2026.4 |
What it is useful for
Fine-tuning and distilling open video generation models so they need fewer denoising steps, and running them with optimized attention backends through a command-line interface or Python API. The README also describes a real-time video generation and editing app built on the framework.1
Run and use notes
- The README's quick start creates a Python 3.12 environment with uv and installs FastVideo with UV_TORCH_BACKEND=cu126 (CUDA 12) or cu130 (CUDA 13). It notes that there is no prebuilt ARM wheel for the FastVideo CUDA kernel, so DGX Spark installs compile it from source.1
Organization context
Provenance and derivatives
The README states that FastVideo learned design ideas from and reused code from Wan-Video, ThunderKittens, DMD2, diffusers, xDiT, vLLM, and SGLang. Checkpoints released through the project, such as the FastH3 models, are distilled from third-party base models (for example MiniMax-H3); those base models are not developed by the Hao AI Lab.1
U.S. eligibility
Eligible · basis: U.S.-governed project
FastVideo is maintained in the Hao AI Lab's GitHub organization. The Hao AI Lab describes itself as a lab at UC San Diego that develops and maintains open-source models, evaluations, and systems. UC San Diego is a public university campus of the University of California in La Jolla, California. Eligibility rests on the maintaining lab; the base video models that FastVideo fine-tunes or distills come from other developers and are not assessed here.1657
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.