FastGen
NVIDIA FastGen: Fast Generation from Diffusion Models
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Check it deeper on your machine
It has changed since its scan on 13 Nov 2026.
As a Claude Code plugin, install it and then ask Claude Code:
/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/NVlabs/FastGen with scanskillsafety
Or from a terminal, clone it and run it:
git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/NVlabs/FastGen'
It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.
Badge for your README
Markdown
[](/listings/nvlabs--fastgen)HTML
<a href="/listings/nvlabs--fastgen"><img src="/api/listings/nvlabs--fastgen/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
**Recommended:** Use the provided Docker container for a consistent environment. See [CONTRIBUTING.md](CONTRIBUTING.md) for Docker setup instructions. Otherwise, create a new [conda](https://www.anaconda.com/docs/getting-started/miniconda/install) environment with `conda create -y -n fastgen python=3.12.3 pip; conda activate fastgen`.
Installation
git clone https://github.com/NVlabs/FastGen.git cd FastGen pip install -e .
On ARM64, PyPI only provides CPU builds of PyTorch, so install the CUDA build first:
pip install "torch>=2.10,<2.11" "torchvision>=0.25,<0.26" --index-url https://download.pytorch.org/whl/cu130 pip install -e .
Credentials (Optional)
For W&B logging, [get your API key](https://wandb.ai/settings) and save it to `credentials/wandb_api.txt` or set the `WANDB_API_KEY` environment variable. Without either of these, W&B will prompt for your API key interactively. For more details, including S3 storage and other environment variables, see [fastgen/configs/README.md](fastgen/configs/README.md#environment-variables).
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
README.md · README · 243 lines
- <h1 align="center">NVIDIA FastGen: Fast Generation from Diffusion Models</h1>
- [](https://docs.nvidia.com/fastgen)
- [](skills/)
- [](LICENSE)
- [](https://pytorch.org/)
- [](https://www.nvidia.com/)
- > **Note:** The current codebase offers early access to new, experimental features. Results reported in the README files were obtained using [FastGen v0.1.0](https://github.com/NVlabs/FastGen/releases/tag/v0.1.0).
- <p align="center">
- <a href="https://jberner.info/">Julius Berner</a> • <a href="https://research.nvidia.com/labs/genair/author/chao-liu/">Chao Liu</a> • <a href="http://latentspace.cc/">Arash Vahdat</a>
- </p>
- <p align="center">
- <a href="https://youtu.be/xEKcP-SwBBY" target="_blank">
- <img src="assets/teaser.png" alt="Watch the video" width=100%/>
- <br/>
- <b>Click here for a demo video</b>
- </a>
- </p>
- FastGen is a PyTorch-based framework for building fast generative models using various distillation and acceleration techniques. It supports:
- - large-scale training with ≥20B parameters.
- - different tasks and modalities, including T2I, I2I, I2V, V2V, multimodal video+audio generation, and causal generation.
- - various distillation methods, including flow maps, distribution matching, parallel decoding distillation, self-forcing, and more.
- ## Repository Structure
- ```
- fastgen/
- ├── fastgen/
- │ ├── callbacks/ # Training callbacks (EMA, profiling, etc.)
- │ ├── configs/ # Configuration system
- │ │ ├── experiments/ # Experiment configs
- │ │ └── methods/ # Method-specific configs
- │ ├── datasets/ # Dataset loaders
- │ ├── methods/ # Trajectory matching, distribution matching, and fine-tuning
- │ ├── networks/ # Neural network architectures
- │ ├── third_party/ # Third-party dependencies
- │ ├── trainer.py # Main training loop
- │ └── utils/ # Utilities (distributed, checkpointing)
- ├── comfyui/ # ComfyUI custom nodes for interactive inference
- ├── scripts/ # Inference and evaluation scripts
- ├── tests/ # Unit tests
- ├── Makefile # Development commands (lint, format, test)
- └── train.py # Main training entry point
- ```
- ## Setup
- **Recommended:** Use the provided Docker container for a consistent environment. See [CONTRIBUTING.md](CONTRIBUTING.md) for Docker setup instructions. Otherwise, create a new [conda](https://www.anaconda.com/docs/getting-started/miniconda/install) environment with `conda create -y -n fastgen python=3.12.3 pip; conda activate fastgen`.
- ### Installation
- ```bash
- git clone https://github.com/NVlabs/FastGen.git
- cd FastGen
- pip install -e .
- ```
- On ARM64, PyPI only provides CPU builds of PyTorch, so install the CUDA build first:
- ```bash
- pip install "torch>=2.10,<2.11" "torchvision>=0.25,<0.26" --index-url https://download.pytorch.org/whl/cu130
- pip install -e .
- ```
- ### Credentials (Optional)
- For W&B logging, [get your API key](https://wandb.ai/settings) and save it to `credentials/wandb_api.txt` or set the `WANDB_API_KEY` environment variable.
- Without either of these, W&B will prompt for your API key interactively.
- For more details, including S3 storage and other environment variables, see [fastgen/configs/README.md](fastgen/configs/README.md#environment-variables).
- ## Quick Start
- Before running the following commands, download the CIFAR-10 dataset and pretrained EDM models:
- ```bash
- python scripts/download_data.py --dataset cifar10
- ```
- For other datasets and models, see [fastgen/networks/README.md](fastgen/networks/README.md) and [fastgen/datasets/README.md](fastgen/datasets/README.md).
- ### Basic Training
- ```bash
- python train.py --config=fastgen/configs/experiments/EDM/config_dmd2_fast.py
- ```
- If you run out-of-memory, try a smaller batch-size, e.g., `dataloader_train.batch_size=32`, which automatically uses gradient accumulation to match the global batch-size.
- **Expected Output:** See the training log for a link to the run on [wandb.ai](https://wandb.ai). Training outputs go to `$FASTGEN_OUTPUT_ROOT/{project}/{group}/{name}/`, where `name` defaults to the config file's name without `.py` (see [Run Directory](fastgen/configs/README.md#run-directory)). With default settings, outputs are organized as follows:
- ```
- FASTGEN_OUTPUT/fastgen/cifar10/config_dmd2_fast/
- ├── checkpoints/ # Model checkpoints in the format {iteration:07d}.pth
- │ ├── 0001000.pth
- │ └── ...
- ├── config.yaml # Resolved configuration for reproducibility
- ├── wandb_id.txt # W&B run ID for resuming
- └── ...
- ```
- ### DDP/FSDP2 Training
- For multi-GPU training, use DDP:
- ```bash
- torchrun --nproc_per_node=8 train.py \
- --config=fastgen/configs/experiments/EDM/config_dmd2_fast.py \
- - trainer.ddp=True log_config.name=test_ddp
- ```
- For large models, use FSDP2 for model sharding by replacing `trainer.ddp=True` with `trainer.fsdp=True`.
- ### Inference
- ```bash
- python scripts/inference/image_model_inference.py --config fastgen/configs/experiments/EDM/config_dmd2_fast.py \
- --classes=10 --prompt_file=scripts/inference/prompts/classes.txt --ckpt_path=FASTGEN_OUTPUT/fastgen/cifar10/config_dmd2_fast/checkpoints/0002000.pth - log_config.name=test_inference
- ```
- For other inference modes and FID evaluations, see [scripts/README.md](scripts/README.md).
- FastGen-distilled models can also be served with:
- - [FlashDreams](scripts/README.md#flashdreams-example-self-forcing-wan-21-t2v) for optimized streaming inference of Self-Forcing models.
- - [vLLM-Omni](https://docs.vllm.ai/projects/vllm-omni/en/latest/api/vllm_omni/diffusion/models/dmd2/) for DMD2 model inference.
- ### Command-Line Overrides
- Override any config parameter using Hydra-style syntax (note the `-` separator):
- ```bash
- python train.py --config=path/to/config.py - key=value nested.key=value
- ```
- Use `+key=value` for a key the config does not define yet; see [fastgen/configs/README.md](fastgen/configs/README.md#command-line-arguments).
- ## Documentation
- The documentation is published at [docs.nvidia.com/fastgen](https://docs.nvidia.com/fastgen). Its pages are
- generated from the READMEs in this repository, plus the authored
- [training and checkpoints](https://docs.nvidia.com/fastgen/training),
- [troubleshooting](https://docs.nvidia.com/fastgen/troubleshooting), and
- [release notes](https://docs.nvidia.com/fastgen/release-notes) pages. See
- [Contributing to the docs](CONTRIBUTING.md#documentation) for how to edit them. The main
- components and their source READMEs are:
- | Component | README | Description |
- |-----------|--------|-------------|
- | **[Methods](https://docs.nvidia.com/fastgen/methods)** | [fastgen/methods/README.md](fastgen/methods/README.md) | Trajectory matching, distribution matching, and fine-tuning methods |
- | **[Networks](https://docs.nvidia.com/fastgen/networks)** | [fastgen/networks/README.md](fastgen/networks/README.md) | Network architectures (Flux, Qwen-Image, Wan, LingBot-world, Cosmos, LTX-2, etc.) and pretrained models |
- | **[Configs](https://docs.nvidia.com/fastgen/configuration)** | [fastgen/configs/README.md](fastgen/configs/README.md) | Configuration system, environment variables, and creating custom configs |
- | **[Datasets](https://docs.nvidia.com/fastgen/datasets)** | [fastgen/datasets/README.md](fastgen/datasets/README.md) | Dataset preparation and WebDataset loaders (image, video, video+audio) |
- | **[Callbacks](https://docs.nvidia.com/fastgen/callbacks)** | [fastgen/callbacks/README.md](fastgen/callbacks/README.md) | Training callbacks (EMA, logging, gradient clipping, etc.) |
- | **[Inference](https://docs.nvidia.com/fastgen/inference)** | [scripts/README.md](scripts/README.md) | Inference modes (T2I, I2I, T2V, I2V, V2V, etc.) and FID evaluation |
- | **[ComfyUI](https://docs.nvidia.com/fastgen/comfyui)** | [comfyui/README.md](comfyui/README.md) | ComfyUI custom nodes for interactive teacher/student inference |
- | **[Skills](https://docs.nvidia.com/fastgen/skills)** | [skills/README.md](skills/README.md) | Agent Skills for coding agents: setup, configuration, data, training, inference, validation, and extending FastGen |
- | **[Third Party](https://docs.nvidia.com/fastgen/third-party)** | [fastgen/third_party/README.md](fastgen/third_party/README.md) | Third-party dependencies (Depth Anything V2, etc.) |
- ## Supported Methods
- | Category | Methods |
- |----------|---------|
- | **Trajectory Matching** | [CM](https://arxiv.org/abs/2303.01469), [ECT](https://arxiv.org/abs/2406.14548), [sCM](https://arxiv.org/abs/2410.11081), [TCM](https://arxiv.org/abs/2410.14895), [MeanFlow](https://arxiv.org/abs/2505.13447), [AYF](https://arxiv.org/abs/2506.14603), [KD](https://arxiv.org/abs/2101.02388), [CausalKD](https://arxiv.org/abs/2412.07772), [PDD](https://arxiv.org/abs/2607.26004) |
- | **Distribution Matching** | [DMD2](https://arxiv.org/abs/2405.14867), [f-Distill](https://arxiv.org/abs/2502.15681), [Phased DMD](https://arxiv.org/abs/2510.27684), [LADD](https://arxiv.org/abs/2403.12015), [AnyFlow](https://arxiv.org/abs/2605.13724), [CausVid](https://arxiv.org/abs/2412.07772), [Self-Forcing](https://arxiv.org/abs/2506.08009), [Causal-Forcing](https://arxiv.org/abs/2602.02214) |
- | **Fine-Tuning** | [SFT](https://arxiv.org/abs/2006.11239), [CausalSFT (diffusion/teacher forcing)](https://arxiv.org/abs/2412.07772) |
- See [fastgen/methods/README.md](fastgen/methods/README.md) for details.
- ## Supported Networks and Data
- FastGen is designed to be **agnostic to the network and data** and you can add your own architectures and datasets (see [fastgen/networks/README.md](fastgen/networks/README.md) and [fastgen/datasets/README.md](fastgen/datasets/README.md)). For reference, we provide the following implementations:
- | Data | Networks |
- |------|----------|
- | **Image** | EDM, EDM2, DiT, SiT, SD 1.5, SDXL, FLUX.1, FLUX.2, Qwen-Image |
- | **Video** | Wan2.1, Wan2.2, Wan Vace, LingBot-world, CogVideoX, Cosmos Predict2, Cosmos 3 |
- | **Video+Audio** | LTX-2, LTX-2.3, LTX-2.5, MiniMax H3 |
- See [fastgen/networks/README.md](fastgen/networks/README.md) for details.
- Not all combinations of methods and networks are currently supported. We provide typical use-cases in our predefined configs in [fastgen/configs/experiments](fastgen/configs/experiments).
- ## Adoption, Collaboration, and Usage
- FastGen is used across NVIDIA and the community to build fast generative models.
- ### Models Distilled Using FastGen
- - **[FastGen-AYF](https://huggingface.co/nvidia/FastGen-AYF)** (distilled to 4-8 steps using AYF from FastGen)
- - **[Qwen-Image-Flash](https://huggingface.co/nvidia/Qwen-Image-Flash)** (distilled to 4 steps using DMD2 from FastGen, [NVIDIA ModelOpt](https://github.com/NVIDIA/Model-Optimizer), and [NVIDIA AutoModel](https://github.com/NVIDIA-NeMo/Automodel))
- - **[Baseten FLUX.2](https://www.baseten.co/blog/faster-image-generation-timestep-distillation-flux2/)** (distilled to 8 steps using DMD2)
- - **[Reka Real-Time Video Generation](https://reka.ai/labs/research/real-time-video-generation)** (distilled to a real-time video generator using FastGen)
- - **[CorrDiff](https://developer.nvidia.com/blog/accelerating-diffusion-models-with-an-open-plug-and-play-offering/)** (distilled to 1 step using ECT)
- - **[cBottle Super-Resolution](https://github.com/NVlabs/cBottle/blob/main/src/cbottle/training/distill_super_resolution.py)** (distilled to 2 steps using ECT)
- ### Projects Using FastGen
- - **[Simplifying and Improving On-Policy Distillation for Few-Step Diffusion](https://zjwfufu.github.io/SMD-Page/)** (built upon FastGen)
- - **[Parametric Trajectory Distillation for Few-Step Video Generation](https://alan-lanfeng.github.io/PTD/)** (built upon FastGen)
- - **[Data-Forcing Distillation: Restoring Diversity and Fidelity in Few-Step Video Generation](https://csy2077.github.io/dfd-project-page/)** (built upon FastGen)
- - **[Parallel Decoding Distillation for Image and Video Generation](https://research.nvidia.com/labs/genair/pdd/)** (built upon FastGen)
- - **[Transition Matching Distillation for Fast Video Generation](https://research.nvidia.com/labs/genair/tmd/)** (built upon FastGen)
- - **[Mode Seeking meets Mean Seeking for Fast Long Video Generation](https://primecai.github.io/mmm/)** (built upon FastGen)
- - **[StreamingEffect: Real-Time Human-Centric Video Effect Generation](https://github.com/showlab/StreamingEffect)** (built upon FastGen)
- - **[CARV: Variance Reduction for Expectations with Diffusion Teachers](https://research.nvidia.com/labs/sil/projects/CARV/)** (used FastGen for the DMD2 experiments)
- - **[Flex-Forcing: Towards a Unified Autoregressive and Bidirectional Video Diffusion Model](https://research.nvidia.com/labs/genair/flex-forcing/)** (built upon FastGen)
- ### FastGen Collaborations
- - **[FastH3 Omni](https://haoailab.com/blogs/fastvideo-omniref-update/)** (Minimax H3 Ref2VA distillation with [FastVideo](https://github.com/hao-ai-lab/FastVideo) and Nuva Lab)
- - **[FastH3 Preview](https://haoailab.com/blogs/fasth3-preview/)** (MiniMax H3 distillation with [FastVideo](https://github.com/hao-ai-lab/FastVideo) and Nuva Lab)
- - **[FlashDreams](https://github.com/NVIDIA/flashdreams)** (optimized serving for FastGen-distilled autoregressive video models)
- - **[NVIDIA Model Optimizer (ModelOpt)](https://github.com/NVIDIA/Model-Optimizer)** (integration of FastGen step-distillation methods)
- ## Contributing
- We welcome contributions! Please see [CONTRIBUTING.md](CONTRIBUTING.md) for details.
- We thank everyone who has helped design, build, and test FastGen!
- - **Core contributors:** Julius Berner, Chao Liu
- - **Other contributors:** Weili Nie, Amirmojtaba Sabour, James Lucas, David Pankratz, Sushil Koundinyan, Sihyun Yu, Willis Ma, Yilun Xu, Shengqu Cai, Xinyin Ma, Yanke Song
- - **Collaborators:** Sophia Zalewski, Wei Xiong, Christian Laforte, Sajad Norouzi, Kaiwen Zheng, Miloš Hašan, Saeed Hadadan, Gene Liu, David Dynerman, Alicia Sui, Grace Lam, Pooya Jannaty, Jan Kautz, and many more.
- - **Project lead:** Arash Vahdat
- ## License
- This project is licensed under the Apache License 2.0 - see [LICENSE](LICENSE) for details. Third-party licenses are documented in [licenses/README.md](licenses/README.md).
- ## Reference
- ```
- @misc{fastgen2026,
- title={NVIDIA FastGen: Fast Generation from Diffusion Models},
- author={Berner, Julius and Liu, Chao and Vahdat, Arash},
- url={https://github.com/NVlabs/FastGen},
- year={2026},
- }
- ```
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.