Skip to content
SPECIFICTION
Get listed

Hallo

fudan-generative-vision.github.io

WindowsLinuxSelf-hosted
Visit fudan-generative-vision.github.io
Spec score 7.6 #14 of 28 in AI Video Lip Sync Tools

What it is

Hallo generates an animated portrait video from a still image and driving speech audio. Its hierarchical audio-driven synthesis is designed to align lip, facial-expression, and pose movement with the audio, while separate inference weights let you adjust those three motion areas. 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; background music is allowed. Inference saves an MP4 to `.cache/output.mp4` by default, with another output path available. Listed requirements are Ubuntu 20.04 or 22.04 and CUDA 12.1, with an A100 named as the tested GPU. Pretrained checkpoints and supporting models are available through Hugging Face. The project lists Linux, Windows, and self-hosted platforms, with community resources for Windows, ComfyUI, a WebUI, Docker, and RunPod. Its repository includes preprocessing and training code for multi-node and multi-GPU work. Hallo is free, open source under MIT, and its training data is English, so driving audio is limited to English.

Spec sheet

Purpose
Hallo is a system for audio-driven visual synthesis that animates portrait images.Read 8 Oct 2026 · github.com
License
GitHub identifies the repository as MIT-licensed.Read 8 Oct 2026 · github.com
Founded
2024Read 28 Sep 2026 · fudan-generative-vision.github.io
Audio requirements
Driving audio must be a WAV file in English, with clear vocals; background music is acceptable.Read 8 Oct 2026 · github.com
Community integrations
The repository links community resources for ComfyUI, a WebUI, Windows, Docker, and RunPod.Read 4 Oct 2026 · github.com
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.Read 8 Oct 2026 · github.com
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.Read 4 Oct 2026 · github.com
Inputs and output
Inference takes a source image and driving audio and saves an animation as an MP4 video by default.Read 8 Oct 2026 · github.com
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.Read 8 Oct 2026 · github.com
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.Read 8 Oct 2026 · github.com
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.Read 8 Oct 2026 · github.com
Motion controls
Inference options include separate pose, face, and lip weights.Read 4 Oct 2026 · github.com
Motion synthesis
Its hierarchical audio-driven synthesis is designed to align lip, expression, and pose motion with audio.Read 4 Oct 2026 · arxiv.org
Output
Inference saves an MP4 video to `.cache/output.mp4` by default, with an option to choose another output path.Read 4 Oct 2026 · github.com
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.Read 4 Oct 2026 · github.com
Research team
The listed authors are affiliated with Fudan University, Baidu, ETH Zurich, and Nanjing University.Read 8 Oct 2026 · github.com
Responsible use
The project notes risks involving deepfakes, privacy, and consent, and calls for ethical guidelines, transparent data use, informed consent, and privacy safeguards.Read 8 Oct 2026 · github.com
System requirements
The listed system requirements are Ubuntu 20.04 or 22.04 and CUDA 12.1; the tested GPU is A100.Read 4 Oct 2026 · github.com
Training
The project provides data-preprocessing and training code, including support for multi-node and multi-GPU training.Read 8 Oct 2026 · github.com
Windows and integrations
The project README lists community contributions for a Windows version, ComfyUI integration, WebUI, and a Docker template.Read 8 Oct 2026 · github.com
Windows support
The repository links a community Windows version and installation guide.Read 4 Oct 2026 · github.com

Strong points and limits

Strong points

  • Adjustable pose, face, and lip weights
  • Outputs MP4 video
  • Pretrained models available on Hugging Face
  • MIT-licensed open source
  • Training code supports multi-GPU work

Limits

  • Driving audio must be English WAV
  • Requires Ubuntu 20.04/22.04 and CUDA 12.1
  • Source image has strict framing requirements
  • A100 is the tested GPU

Who it suits

It suits developers creating talking-portrait videos from controlled image and audio inputs. Users need the listed Ubuntu/CUDA setup or community integrations and English driving audio.

Verdict

Hallo offers configurable audio-driven portrait animation with an explicit input format and output path. Its deployment requirements and English-audio limitation are important constraints for prospective users.

Where Specifiction places it

Alternatives to HalloAll alternatives →