Hallo
What it is
Hallo is free, open-source software that turns a still portrait and spoken audio into an animated MP4 video. Its audio-driven synthesis coordinates lip movement, facial expression, and pose with the supplied speech. Separate inference weights let users adjust pose, face, and lip motion. The source image should be square, show a forward-facing face occupying 50%–70% of the frame, and have less than 30 degrees of rotation. Driving audio must be a clear-vocal English WAV file, although background music is acceptable. By default, output is saved as `.cache/output.mp4`, with another path available. The listed requirements are Ubuntu 20.04 or 22.04 and CUDA 12.1, with an A100 listed as the tested GPU. Pretrained models come from Hugging Face. Community resources cover Windows, ComfyUI, WebUI, Docker, and RunPod. The project uses an MIT license.
For and against
For
- Free and open source under an MIT license
- Separate controls for pose, face, and lips
- Exports MP4 video by default
- Pretrained models available through Hugging Face
- Training and preprocessing code provided
Against
- Driving audio must be English WAV
- Requires Ubuntu and CUDA 12.1
- A100 is the listed tested GPU
- Portrait images have strict framing requirements
The verdict
Hallo provides a focused image-to-speaking-portrait workflow with adjustable motion controls and open-source code. Its English-only audio requirement and CUDA-based setup are important constraints.
Who it suits
Hallo suits users creating portrait animations from compatible images and English speech recordings. It is aimed at Linux, Windows, or self-hosted setups that meet the listed system requirements.
On the record
- Audio requirements
- Driving audio must be a WAV file in English, with clear vocals; background music is acceptable.github.com · read 8 Oct 2026
- Community integrations
- The repository links community resources for ComfyUI, a WebUI, Windows, Docker, and RunPod.github.com · read 4 Oct 2026
- Image requirements
- The source image should be square, show the face as 50%–70% of the image, and have a forward-facing head with less than 30° rotation.github.com · read 8 Oct 2026
- Input image requirements
- The source image should be square, with a forward-facing face occupying 50%–70% of the image and rotated less than 30 degrees.github.com · read 4 Oct 2026
- Inputs and output
- Inference takes a source image and driving audio and saves an animation as an MP4 video by default.github.com · read 8 Oct 2026
- Installation
- The project lists Ubuntu 20.04 or 22.04 and CUDA 12.1 as system requirements, and says it tested the software on A100 GPUs.github.com · read 8 Oct 2026
- Language limitation
- The README says the model's training datasets are only in English and lists improving Mandarin performance as a roadmap item with no estimated date.github.com · read 8 Oct 2026
- License
- GitHub identifies the repository as MIT-licensed.github.com · read 8 Oct 2026
- Model downloads
- Pretrained models for inference are available from the project's Hugging Face repository, with separate downloads also listed for components such as wav2vec and Stable Diffusion 1.5.github.com · read 8 Oct 2026
- Motion controls
- Inference options include separate pose, face, and lip weights.github.com · read 4 Oct 2026
- Motion synthesis
- Its hierarchical audio-driven synthesis is designed to align lip, expression, and pose motion with audio.arxiv.org · read 4 Oct 2026
- Output
- Inference saves an MP4 video to `.cache/output.mp4` by default, with an option to choose another output path.github.com · read 4 Oct 2026
- Privacy and misuse
- The project identifies deepfake misuse and image and voice privacy and consent as social risks, and calls for responsible use, informed consent, and privacy safeguards.github.com · read 4 Oct 2026
- Purpose
- Hallo is a system for audio-driven visual synthesis that animates portrait images.github.com · read 8 Oct 2026
- Research team
- The listed authors are affiliated with Fudan University, Baidu, ETH Zurich, and Nanjing University.github.com · read 8 Oct 2026
- Responsible use
- The project notes risks involving deepfakes, privacy, and consent, and calls for ethical guidelines, transparent data use, informed consent, and privacy safeguards.github.com · read 8 Oct 2026
- System requirements
- The listed system requirements are Ubuntu 20.04 or 22.04 and CUDA 12.1; the tested GPU is A100.github.com · read 4 Oct 2026
- Training
- The project provides data-preprocessing and training code, including support for multi-node and multi-GPU training.github.com · read 8 Oct 2026
