Wan2.1 VACE

Wan2.1 VACE

Wan2.1 all-in-one create/edit member: reference images, V2V, and mask/control in one native Wan assembly.

Wan2.1 VACE is the all-in-one video creation and editing member of the Wan2.1 family: reference-image guidance, video-to-video editing, and mask/control-conditioned generation in one checkpoint. WorldFoundry assembles it from canonical Wan roles; image references, source video/mask conditions, and precomputed VACE context tensors are all supported without an official-runtime backend.

Official paper teaser: VACE as an all-in-one video creation and editing model on the Wan stack.
Official paper teaser: VACE as an all-in-one video creation and editing model on the Wan stack.

Build your run

Pipeline
Wan2p1VACEPipelineworldfoundry.pipelines.wan.pipeline_wan_vace
TaskInteractive Video interactive-video
Environmentworldfoundry-unified-cu128
DeviceCUDA 12.8
Runnerworldfoundry.pipeline
Loadingfrom_pretrained
Invocationnative_or_standard_invocation
worldfoundry-eval run \
  wan2.1-vace \
  --pipeline.task-profile interactive-video \
  --json

What it is for

Start on wan2.1-vace. Launch worldfoundry-eval run wan2.1-vace. Use this card for reference-image create/edit, V2V, and mask/control — not the plain Wan2.1 T2V backbone.

VBench is the listed quality suite.

Runtime contract

Required prompt; optional image / video. Native defaults: 1280×720, 81 frames, 50 steps, guidance 5.0. Artifact: wan2.1-vace.mp4. Image references, source video/mask conditions, and precomputed VACE context tensors are supported without an official-runtime backend.

Running in WorldFoundry

Unified worldfoundry-unified-cu128 (Python 3.11, CUDA 12.8). Stage Wan-AI/Wan2.1-VACE-14B or Wan-AI/Wan2.1-VACE-1.3B-diffusers. Status: Runner parity pending for full checkpoint visual parity. Prepare masks and references before launch.