VASA-1
Research framework for lifelike real-time talking faces from a single image and speech audio.
About
VASA-1, from Microsoft Research Asia (arXiv 2404.10667, April 2024), generates lifelike talking faces from one static portrait and a speech clip. Its core is a diffusion-based model that produces holistic facial dynamics and head movements inside a disentangled face latent space, which lets it separate identity, pose and expression — the famous demos include the Mona Lisa rapping. It renders 512x512 video at up to 40 frames per second, fast enough for real-time avatar interaction.
Microsoft has kept it strictly a research demonstration: the company said it has no plans to release a demo, API, product or implementation details until responsible-use questions are settled. So VASA-1 matters to the catalog as the influential research milestone that set the bar for audio-driven avatars — OmniHuman, Hallo3 and their peers all followed — rather than as something you can use today.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- Paper-reported (arXiv 2404.10667): outperforms MakeItTalk, Audio2Head and SadTalker on audio-lip synchronization, pose alignment and Frechet Video Distance
- No independent third-party benchmarks published
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 4/10
Scored 4/10 (TIER metrics, metrics-v1). Evidence: peer-style technical report (arXiv 2404.10667) with comparisons against MakeItTalk/Audio2Head/SadTalker; official Microsoft Research project page; wide press coverage. Strengths: genuinely influential research, strong reported results. Docked for: (1) no weights, API, demo or code — Microsoft explicitly declined to release it; (2) no maintenance or updates since 2024; (3) evals are author-reported. License: proprietary / unreleased.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 5.0/10
- Maintenance · recency, releases, issue load
- 3.0/10
- Openness · license + docs
- 1.0/10
- Manual review · sanity read + red-flag check
- 5.0/10
- · Microsoft Research Asia, arXiv 2404.10667 (Apr 2024); 512x512 @ up to 40fps
- · Microsoft explicitly declined to release demo/API/weights/code
- · Influential research milestone; evals are author-reported only