Repolode
Submit
aigc-apps

aigc-apps/VideoX-Fun

A video generation framework for Diffusion Transformer models that creates videos at flexible resolutions and from images, with training support for baseline and LoRA models.

Open-source alternative to:RunwayPika

View on GitHub
Stars
2,311
Forks
197
Open issues
132
Last commit
2 days ago

What is it?

VideoX-Fun is a pipeline for generating AI images and videos with Diffusion Transformer models. It runs inference from pre-trained base models to produce videos at different resolutions, durations and frame rates, and it also lets users train their own base and LoRA models for specific style transformations.

The project covers several model families, including CogVideoX-Fun, Wan2.1, Wan2.2 and their Fun variants, with tasks such as text-to-video, image-to-video, video-to-video and controlled generation using Canny, Pose or Depth conditions. Inference is available through Python scripts, a Gradio web UI, or ComfyUI nodes, and GPU memory options such as CPU offload and float8 quantization are provided for consumer GPUs.

Training scripts cover data preprocessing from JSON-indexed image and video datasets, plus baseline, LoRA and control training for each supported model. Multi-GPU inference is supported through xfuser with configurable Ulysses and ring parallelism.

Key features

  • 01Generates videos at multiple resolutions, durations and frame rates from pre-trained models.
  • 02Supports text-to-video, image-to-video, video-to-video and controlled generation.
  • 03Trains custom baseline and LoRA models on your own image and video datasets.
  • 04Offers Python scripts, a Gradio web UI and ComfyUI integration for inference.
  • 05Provides GPU memory modes such as CPU offload and float8 quantization.
  • 06Supports multi-GPU inference with xfuser.

Use cases

  • Creating short AI videos from a still image or text prompt.
  • Training a LoRA to apply a specific visual style to generated video.
  • Running controlled video generation guided by pose or depth maps.
  • Batch-generating videos on rented or local GPUs with scripted parameters.

Quick start

  1. 1

    Pull the official Docker image

    docker pull mybigpai-public-registry.cn-beijing.cr.aliyuncs.com/easycv/torch_cuda:cogvideox_fun
  2. 2

    Start a container from the image with GPU access

    docker run -it -p 7860:7860 --network host --gpus all --security-opt seccomp:unconfined --shm-size 200g mybigpai-public-registry.cn-beijing.cr.aliyuncs.com/easycv/torch_cuda:cogvideox_fun
  3. 3

    Clone the VideoX-Fun repository inside the container

    git clone https://github.com/aigc-apps/VideoX-Fun.git

Things to know

  • Requires an NVIDIA GPU with substantial VRAM, and the README lists 12GB as the smallest tested card.
  • Needs about 60GB of free disk space for model weights.
  • Multi-GPU inference requires extra installs (xfuser and yunchang) and careful parallel degree settings.

Topics

#video-generation#diffusion-transformer#text-to-video#image-to-video#lora#wan#cogvideox#pytorch

FAQ

What is VideoX-Fun?
A video generation framework for Diffusion Transformer models that creates videos at flexible resolutions and from images, with training support for baseline and LoRA models. VideoX-Fun is a pipeline for generating AI images and videos with Diffusion Transformer models. It runs inference from pre-trained base models to produce videos at different resolutions, durations and frame rates, and it also lets users train their own base and LoRA models for specific style transformations.
Is VideoX-Fun free to use? What license does it have?
VideoX-Fun is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is VideoX-Fun still maintained?
Actively maintained. The last commit to VideoX-Fun was on Oct 20, 2026.
What can I use VideoX-Fun for?
Creating short AI videos from a still image or text prompt. Training a LoRA to apply a specific visual style to generated video. Running controlled video generation guided by pose or depth maps. Batch-generating videos on rented or local GPUs with scripted parameters.
How do I install VideoX-Fun?
Pull the official Docker image: docker pull mybigpai-public-registry.cn-beijing.cr.aliyuncs.com/easycv/torch_cuda:cogvideox_fun Start a container from the image with GPU access: docker run -it -p 7860:7860 --network host --gpus all --security-opt seccomp:unconfined --shm-size 200g mybigpai-public-registry.cn-beijing.cr.aliyuncs.com/easycv/torch_cuda:cogvideox_fun Clone the VideoX-Fun repository inside the container: git clone https://github.com/aigc-apps/VideoX-Fun.git
Where can I get VideoX-Fun?
The source code and installation instructions are on GitHub: https://github.com/aigc-apps/VideoX-Fun

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects