Wan video-to-video editing uses an existing 2–10 second MP4 or MOV plus an instruction to create a changed version. It is suited to restyling, localized scene changes, and preserving an existing camera path while altering the look. The current WanVideoFree integration uses the hosted Wan 2.7 edit-video endpoint.
State the transformation first, then the invariants:
Transform the entire scene into a soft watercolor painting on textured paper. Preserve the subject, timing, camera movement, and composition. Keep the original audio.
Avoid asking for several incompatible transformations in one pass. If identity or product geometry must remain exact, name that constraint directly and review the result before commercial use.
Wan 2.7 image-to-video also accepts a short video URL to continue a clip. Use continuation when the scene should advance beyond the last frame. Use edit-video when the timing and camera path should stay recognizable while the visual content changes.
The service checks high-risk prompt categories and rejects private/internal URLs before provider spend. The provider safety checker remains enabled. Eligible credits are returned for a submit failure, provider failure, or server-settled timeout; a completed output that does not match personal taste is not automatically a failed task.
Source: Wan 2.7 edit-video API schema. Last checked August 12, 2026.
Yes, the Wan 2.7 image-to-video endpoint accepts a 2–10 second video for continuation. The edit-video endpoint is better when you want to restyle or alter the supplied clip rather than simply continue it.
The edit endpoint exposes an audio setting that can preserve the original audio. The result still needs review because generated visual timing or provider processing can vary.
The inference provider must be able to fetch the file. WanVideoFree rejects localhost, private-network, credential-bearing, and non-HTTPS URLs before any paid call to prevent server-side request abuse.