VGen
ali-vilab/VGenReviewed by AI
Checked by RepoPicks Radar on
What is VGen?
VGen is the research codebase of Alibaba's Tongyi Lab. It gathers several video generation methods in one place — I2VGen-XL (photo to video), HiGen and TF-T2V (text to video), DreamVideo (your own subject or motion), VideoLCM and others — and you drive each one by editing a YAML file and running inference.py, or by opening a local Gradio page. It can also train and fine-tune models. For a first look without installing anything, the authors keep browser demos on HuggingFace and ModelScope. The repository has no license file.
VGen is free and can replace Runway, LTX Studio, and Google Veo 3.
How easy is it to install VGen?
A · download and open · Web
- Linux D
- Web A
The best way in is the browser: the README links demos of I2VGen-XL run by the authors on HuggingFace Spaces and ModelScope, where you upload a picture and get a video without installing anything. Doing it on your own machine is a D and really a Linux job: a conda environment with Python 3.8, PyTorch 1.12 built for CUDA 11.3, ffmpeg installed with apt-get, model files pulled from ModelScope and a YAML config per model.
VGen on GitHub
- Stars
- 3,152on Nov 10, 2026
- Downloads
- —all releases
- Growth
- —no gain in the last reading
Stars over time
Facts
- Runs on
- Linux, Web
- You need to install
- Python, NVIDIA graphics card, FFmpeg
- To build from source
- Python, Graphics card, FFmpeg
- Disk space
- 310 MB–870 MB · estimate
- Accounts or keys
- None
- Cost
- Free · No price. There is also no license file, so the terms for using the code and the videos it makes are not stated.
- License
- No license (No license)
- Can replace
- Runway, LTX Studio, Veo 3
- Last change on GitHub
- Jan 10, 2025
VGen risk signals (6)
- AttentionDependencies with known vulnerabilities
Pinned dependencies with known vulnerabilities (OSV): transformers: GHSA-282v-666c-3fvg, GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm; opencv-python: GHSA-qr4w-53vh-m672, PYSEC-2023-183; opencv-python-headless: GHSA-jh2j-j4j9-crg3, PYSEC-2023-184; pytorch-lightning: GHSA-2vj5-px25-gjrp, GHSA-4cv3-v7pv-rfhf, GHSA-75m9-98v2-hjpm; gradio: GHSA-26jh-r8g2-6fpr, GHSA-279j-x4gx-hfrh, GHSA-34rf-p3r3-58x2
automatic check
- InfoNo license detected
No license detected by GitHub, the license file, the LICENSES folder or the winget manifest
automatic check
- Infomodels and packages come from outside github
The README downloads every checkpoint from ModelScope (`snapshot_download('damo/I2VGen-XL', ...)`, `iic/HiGen`, `iic/tf-t2v`, `iic/InstructVideo`) and installs the requirements from a mirror: "pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple".
AI review
- Infostopped being updated
Last commit 2025-01-10, 664 days before the check, with 116 open issues, and the pinned versions are old: torch 1.12 with CUDA 11.3, xformers 0.0.13, pytorch-lightning 1.4.2, gradio 3.39 and a dated triton build.
AI review
- Infoplanned features never shipped
The TODO list still has unchecked items — "Release the code and pretrained models of HumanDiff", "Release models optimized specifically for the human body and faces" — and section "(8) Other methods" says only "In preparation!!".
AI review
- Infoauthor states weak spots of the model
README: "we find that the current model performs inadequately on anime images and images with a black background due to the lack of relevant training data", and the super-resolution model "only supports 32-frame input".
AI review
Evidence is shown in its original language.
What Radar could not check about VGen
- Whether the HuggingFace and ModelScope demos are still online and how long the queue is
- How much graphics memory each model needs, which the README never states
- The real disk space taken by the checkpoints downloaded from ModelScope
- Whether the Python 3.8 and CUDA 11.3 environment still installs on a current machine
- What you are allowed to do with the generated videos, since there is no license file