Wan 2.7 Video Edit
What is Wan 2.7 Video Edit?
Wan 2.7 Video Edit is the dedicated video-to-video model in the Wan 2.7 family. It transforms an existing video using a text prompt and optional reference images. The input clip runs 2 to 10 seconds; output renders at 720P or 1080P. Up to four reference images can guide style or content. Audio can pass through unchanged via audio_setting: "origin", or the model can decide automatically. Use it for style transfer, background swaps, lighting changes, character restyling, and brand restyling on existing footage.
Key features of Wan 2.7 Video Edit
Four features cover what the model exposes.
Video-to-video editing
Pass a 2–10 second source clip via `video_url`. The model preserves source motion while transforming appearance based on the prompt and any reference images.
Up to 4 reference images for style
Pass `reference_image_urls` to anchor style or content. Useful for outfit replacement, prop placement, and brand-aligned restyles.
Audio passthrough or auto
Set `audio_setting: "origin"` to keep the source audio unchanged. Default `auto` lets the model decide based on the prompt. Useful for preserving voiceover while restyling visuals.
Optional duration and aspect override
Omit `duration` and `ratio` to keep the input length and aspect, or force a value when the output needs a specific shape. Useful for repurposing clips into multiple cuts.
Best for
Style transfer
Apply artistic styles to existing video footage in a single call.
Background swaps
Replace the background while preserving subject motion from the source clip.
Brand adaptation
Restyle source footage to match brand colour and lighting.
Seasonal transformations
Day-to-night, summer-to-winter, era shifts on the same source.
Creative exploration
Run multiple edit directions on the same source clip.
Pipeline plug-in
Drop into a content pipeline that already produces source video and needs a styling pass.
Limitations
Source video is capped at 2 to 10 seconds and 100MB. Output duration is 2 to 10 seconds (omit to keep input length). Reference images cap at four, max 20MB each. Resolution is 720P or 1080P only. The model preserves source motion, so structural changes that require entirely new motion (e.g. replacing a walking subject with a running one) are out of scope; use Wan 2.7's reference-to-video mode for that.
API examples
Call Wan 2.7 Video Edit from any language by POSTing to /v1/tasks. Full parameter docs live at docs.unifically.com/models/video/alibaba/wan-2.7-video-edit.
curl -X POST https://api.unifically.com/v1/tasks \
-H "Content-Type: application/json" \
-H "Authorization: Bearer YOUR_API_KEY" \
-d '{
"model": "alibaba/wan-2.7-video-edit",
"input": {
"prompt": "Convert the entire scene to a claymation style",
"video_url": "https://example.com/scene.mp4",
"resolution": "720P"
}
}'
Successful submission returns a task_id. Poll GET /v1/tasks/<task_id> or set a callback_url on the request to receive the finished result.
FAQs
People also ask
Wan 2.7 Video Edit (alibaba/wan-2.7-video-edit) is the dedicated video-to-video model in the Wan 2.7 family. It transforms an existing video using a text prompt and optional reference images. Source clips are 2 to 10 seconds; output runs at 720P or 1080P.
The base Wan 2.7 generates new videos from text, image, or reference inputs. Wan 2.7 Video Edit takes an existing video as input via video_url and modifies it with a prompt and up to four reference images. Different model ID, different surface.
Up to four. Useful for style transfer, outfit replacement, prop placement, and other content swaps that need a visual anchor.
Style transfer (apply a look across frames), background swap, lighting and colour shifts, character or outfit restyling, and seasonal or thematic transformations of existing footage.
Yes. Set audio_setting to origin to keep the source audio unchanged. The default is auto, which lets the model decide based on the prompt.
Both are optional. Omit duration to keep the input length (2 to 10 seconds). Omit ratio to keep the input aspect ratio. Force a different value when needed.