# GAIR-NLP/daVinci-MagiHuman

Repository: https://github.com/GAIR-NLP/daVinci-MagiHuman
Canonical: https://ross.abutalabs.com/products/davinci-magihuman
Language: Python
License Family: other
Last push: 2026-04-11T09:09:32+00:00

## Health v2 (maintenance only)
Score: 49/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 76, release rhythm 35, longevity 11
- inputs: {"age_days": 164, "days_push": 144, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, young, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2113, forks 214 (observed 2026-08-28T04:06:14.638596+00:00)

## What it is
daVinci-MagiHuman is an open-source 15B-parameter single-stream transformer foundation model that jointly generates synchronized audio and video from text prompts, with a focus on human-centric performance. It ships the full model stack (base, distilled, super-resolution) plus inference code in Python/PyTorch.

## Use cases
- generate talking-head videos with synchronized speech from text
- create multilingual avatar videos with expressive faces and body motion
- fast audio-video generation on a single GPU
- generate 1080p human videos with super-resolution
- research on unified multimodal generation architectures
- generate videos in Chinese, English, Japanese, Korean, German, or French

## When to choose
- you need fast, high-quality audio-video generation of humans
- you want a fully open model stack including distilled and super-resolution models
- you need multilingual speech with accurate lip and body synchronization
- you have an H100-class GPU and want state-of-the-art human-centric video generation

## When to avoid
- you lack a high-end GPU (inference targets H100-class hardware)
- you need non-human or scene-dominant video generation
- you need a lightweight CPU-only solution
- you require a permissively usable model but your project forbids Apache-2.0 terms

## Facets
- artifact type: library
- maturity: active
- function: video-processing, audio-processing, deep-learning, llm-inference, machine-learning
- domain: deep-learning, artificial-intelligence, media
- platform: python
- tags: video-generation, audio-video-generation, talking-head, single-stream-transformer, foundation-model, text-to-video, multilingual, huggingface, video, audio, gpu, linux

## Member repositories
- GAIR-NLP/daVinci-MagiHuman (main) score 49

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:14.638596+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:53:30.291749+00:00, confidence not recorded.
  - readme: https://github.com/GAIR-NLP/daVinci-MagiHuman (fetched 2026-08-28T04:06:14.638596+00:00, sha c4a55204ed67)
- Data as of 2026-08-30T08:39:29.467469+00:00.