Wan 2.1 VACE
Alibaba's all-in-one open model for video creation AND editing: reference-to-video, repainting, masked edits, and spatio-temporal extension.
About
Wan2.1-VACE (Video All-in-one Creation and Editing), released May 2025 as part of Alibaba's Wan2.1 family, was presented as the first open-source model unifying video generation and editing in a single framework. The core is the Video Condition Unit (VCU), a unified interface for text, image, video, and mask inputs, plus a Context Adapter that encodes temporal and spatial task requirements. From that one model you get reference-to-video (animate a character from stills), video-to-video repainting, masked region editing (add, alter, or remove things in a selected area), first-last-frame interpolation, inpainting/outpainting, and spatio-temporal boundary extension.
It ships in 1.3B and 14B sizes on the Wan2.1 base, with weights on Hugging Face and ModelScope and code in the Wan-Video GitHub org. The official repos are Apache-2.0, though one third-party Hugging Face mirror labels its re-upload under a custom WAN license, so users should pull weights from official sources. An ICCV 2025 paper backs the method, but there are no published benchmark scores, and as a mid-2025 release it now sits alongside the newer Wan 2.2 and 2.5/2.6 generations.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
No published benchmarks
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (metrics tier, metrics-v1). Evidence: Alibaba Wan2.1 series, VACE announced mid-May 2025 (UK Tech News 2025-05-16, CMOtech), ICCV 2025 paper; official GitHub org Wan-Video (repos Apache-2.0), VACE-1.3B/14B weights on HF/ModelScope. Strengths: genuinely unified generation-plus-editing in one open model, permissive license, strong backing. Docked for: (1) no published benchmarks; (2) license anomaly: a third-party HF mirror claims a custom WAN license, so source of weights matters. License: Apache-2.0 per official Wan-Video repos; anomaly flagged.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 8.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Alibaba Wan2.1 family; announced mid-May 2025; ICCV 2025 paper
- · Video Condition Unit (VCU) unifies text/image/video/mask inputs; 1.3B and 14B sizes
- · Weights on HF/ModelScope; code in Wan-Video/Wan2.1
- · No published benchmarks