#3131st of 75 in Image and videoFor image, video and design, video and animation
fastvideo
Post-training and inference framework for accelerated video diffusion models
- Stars
- 4.8k
- License
- Apache-2.0
- Last commit
- Nov 2026
- Last release
- Jun 2026
- Language
- Python
3,368 on 22 Apr 20264,811 on 11 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
FastVideo is a Python framework for fine-tuning, distilling and running open video diffusion transformers. It covers full and LoRA finetuning, DMD2 and sparse distillation with Video Sparse Attention (claimed >50x denoising speedup), and inference through a CLI and a VideoGenerator Python API with sequence parallelism. It also ships Dreamverse, a realtime video generation web app, and supports NVIDIA, Apple Silicon (MLX), and AMD ROCm.
Who it is for: Engineers fine-tuning, distilling or serving open video diffusion models
Strengths
- Covers training and inference: LoRA, full finetune, DMD2, Self-Forcing, FSDP2, sequence parallelism
- Runs on NVIDIA CUDA 12/13, DGX Spark, Apple Silicon via MLX, and AMD ROCm
- Publishes distilled checkpoints (FastWan, FastH3) on Hugging Face with cookbook recipes
- Installs with uv pip; includes Dreamverse realtime app with a live demo
Weaknesses
- Sparse-attention speedups depend on VSA kernels, which need a separate install step
- Multi-GPU use on low-RAM hosts like Colab risks OOM; README advises num_gpus=1
- Model and hardware support varies; readers must check the support matrix
- No Docker or compose files detected for the core framework
What it needs
- GPU required
- Docker
- Needs PyTorch, uv, VSA attention kernels
- Models: Wan2.1, FastWan, FastH3 (MiniMax-H3), Kandinsky 6, PAN 2
What changed
From the release notes. No flag means the note does not say so, not that the upgrade is safe.
Also in Image and video
See all 75| Rank | Project | Score |
|---|---|---|
| 1 | frigateLocal NVR with realtime object detection for IP cameras | 93 out of 100 |
| 2 | MoneyPrinterTurboGenerates short videos from a topic, with script, footage, voiceover and subtitles | 88 out of 100 |
| 3 | toonflow-appInfinite-canvas tool for AI short drama and comic video production | 87 out of 100 |
| #4 | InvokeAIWeb UI for diffusion image generation with inpainting canvas and node workflows | 86 out of 100 |
| #5 | ComfyUINode-graph editor for running diffusion image, video, audio and 3D workflows | 85 out of 100 |