Run, train, and accelerate video generation models with FastVideo.
FastVideo is an inference and post-training framework for diffusion models. It features an end-to-end unified pipeline for accelerating diffusion models, starting from data preprocessing to model training, finetuning, distillation, and inference. FastVideo is designed to be modular and extensible, allowing users to easily add new optimizations and techniques. Whether it is training-free optimizations or post-training optimizations, FastVideo has you covered.
Start with a FastVideo modelFastH3 8-Step V2Configure a recipe →Recent news
All announcements →Quick start
Generate your first video with the Python API.
Installation
Set up FastVideo on NVIDIA GPUs or Apple Silicon.
Cookbook
Choose a model recipe, customize its settings, and download a serving configuration.
Training and distillation
Prepare data, fine-tune models, and build faster generators.
Key Features
FastVideo has the following features:
- End-to-end post-training support for bidirectional and autoregressive models
- Full finetuning and LoRA finetuning for state-of-the-art open video DiTs
- Data preprocessing pipeline for video, image, and text data
- Distribution Matching Distillation (DMD2) stepwise distillation
- Sparse attention with Video Sparse Attention
- Sparse distillation to achieve >50x denoising speedup
- Attn-QAT training for quantization-aware post-training
- Causal distillation through Self-Forcing
- Scalable training with FSDP2, sequence parallelism, and selective activation checkpointing
- See the training overview for the full training workflow
- State-of-the-art performance optimizations for inference
- Sequence parallelism for distributed inference
- Multiple state-of-the-art attention backends
- User-friendly CLI and Python API
- See the support matrix for supported models and optimizations for the full list
- Realtime video generation and editing
- Dreamverse: stream and “vibe direct” video in realtime (live demo)
Projects and research
Explore work using FastVideo’s code or the attention research behind it.
- SGLang: SGLang’s diffusion inference functionality is based on a fork of FastVideo on Sept. 24, 2025.
- DanceGRPO: A unified framework to adapt Group Relative Policy Optimization (GRPO) to visual generation paradigms. Code based on FastVideo.
- SRPO: A method to directly align the full diffusion trajectory with fine-grained human preference. Code based on FastVideo.
- DCM: Dual-expert consistency model for efficient and high-quality video generation. Code based on FastVideo.
- HY-WorldPlay: An action-conditioned world model model trained using FastVideo framework.
- Hunyuan Video 1.5: A leading lightweight video generation model, where they proposed SSTA based on Sliding Tile Attention.
- Kandinsky-5.0: A family of diffusion models for video & image generation, where their NABLA attention includes a Sliding Tile Attention branch.
- LongCat Video: A foundational video generation model with 13.6B parameters with block-sparse attention similar to Video Sparse Attention.