DaGAN
harlanhong/CVPR2022-DaGANReviewed by AI
Checked by RepoPicks Radar on
What is DaGAN?
DaGAN is the research code behind a CVPR 2022 paper on talking head video generation. You give it one photo of a face and a short video of someone talking; it produces a video where the face in the photo copies that head motion and expression. It is aimed at researchers and curious technical users: you clone the repository, install Python packages, download a model checkpoint from a OneDrive link, crop the inputs with ffmpeg and run a command. The README also links a free Hugging Face demo page that needs nothing installed.
DaGAN is free and open source (MIT) and can replace D-ID, HeyGen, and Viggle.
How easy is it to install DaGAN?
A · download and open · Web
- Linux D
- Web A
On the web the README links a ready Hugging Face demo page, so you only open it and upload a photo and a video. Running it yourself is rated for Linux, the system the README recommends: you install Python packages, build the face-alignment library, fetch a checkpoint from an outside OneDrive link, crop the video with ffmpeg and run a command line, and a GPU is advised.
DaGAN on GitHub
- Stars
- 997on 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, Graphics card, FFmpeg
- To build from source
- Python, Graphics card, FFmpeg
- Disk space
- 900 MB–4.2 GB · estimate
- Accounts or keys
- None
- Cost
- Free · No paid service is involved. The LICENSE.md file is Creative Commons Attribution-NonCommercial 4.0, so commercial use is not granted; the README asks you to email the author about commercial use.
- License
- MIT (Permissive)
- Last change on GitHub
- Dec 4, 2023
DaGAN risk signals (5)
- AttentionDependencies with known vulnerabilities
Pinned dependencies with known vulnerabilities (OSV): certifi: GHSA-248v-346w-9cwc, GHSA-43fp-rhv2-5gv8, GHSA-6q48-6c7w-9gw2; fonttools: GHSA-6673-4983-2vx5, GHSA-768j-98cg-p3fv, PYSEC-2026-1388; grpcio: GHSA-496j-2rq6-j6cc, PYSEC-2026-1424; joblib: GHSA-6hrg-qmvc-2xh8, PYSEC-2022-288; markdown: GHSA-5wmx-573v-2qwq, PYSEC-2026-89; numpy: GHSA-fpfv-jqm9-f5jm; pillow: GHSA-22hg-r758-rrw9, GHSA-24rc-8hj2-v98c, GHSA-254f-3936-288h; protobuf: GHSA-7gcm-g887-7qv7, GHSA-8gq9-2x98-w8hf, GHSA-8qvm-5x2c-j2w7; scikit-learn: GHSA-jw8x-6495-233v, PYSEC-2024-110; scipy: PYSEC-2023-102, PYSEC-2023-114; tqdm: GHSA-g7vv-2v7x-gj9p, PYSEC-2026-1976; werkzeug: GHSA-29vq-49wr-vm6x, GHSA-2g68-c3qc-8985, GHSA-87hc-h4r5-73f7; zipp: GHSA-jfmj-5v4g-7637, PYSEC-2026-2074
automatic check
- Attentionlicense file noncommercial
The LICENSE.md file in the repository is "Attribution-NonCommercial 4.0 International" (Creative Commons), which grants use "for NonCommercial purposes only", while GitHub reports no clear license (NOASSERTION). The README says: "If you have any question or collaboration need (research purpose or commercial purpose), please email [email protected]".
AI review
- Attentiongenerates deepfake faces
The project makes a real person's face in a photo speak and move with someone else's head motion. Repository topics include "deepfake" and "face-reenactment", and the description says "Talking Head Video Generation". Videos of a person who never said those words can be made this way, and several countries restrict images of a person made without consent.
AI review
- Infomodel from outside file sharing
The model files are not in the repository: the README points to personal OneDrive folders at hkustconnect-my.sharepoint.com for "The pre-trained checkpoint of face depth network and our DaGAN checkpoints", so the download is not versioned with the code.
AI review
- Infono longer updated
The last commit is from 2023-12-04 and there are no GitHub releases. The README points to a newer project by the same author, MCNet (ICCV2023), as the convenient option.
AI review
Evidence is shown in its original language.
What Radar could not check about DaGAN
- Whether the Hugging Face demo page linked in the README is still online and working
- Whether the pinned package versions in requirements.txt still install on a current Python and GPU driver (the file pins torchaudio==0.10.1+rocm4.1)
- How well the result looks on your own photo, and how long one video takes on a machine without an NVIDIA GPU
- Real disk use after downloading the depth model and the DaGAN checkpoints from the OneDrive links
- Whether the OneDrive links still hold the files