Use a single-frame character_mask when the character region should stay fixed throughout the output: Wan Animate repeats that mask frame for every output frame. Use a multi-frame mask when the region needs to change over time; the node reads input videos, including the mask, from video_frame_offset. Before running, check that the offset and available mask frames cover the interval you intend to process. These are input semantics, not automatic character tracking or silhouette repair.
Decide and align the mask interval
Use the character region, not just the fact that the source video moves, to choose the mask. A locked-off shot with a stationary performer may suit one frame; a moving performer or changing outline calls for a sequence if the mask must follow those changes. This is an editorial decision about the desired coverage, not a promise that a mask will improve generation.
- In the
WanAnimateToVideonode, identify the intended outputlengthand currentvideo_frame_offset. For chunked work, note which portion of the source video this offset is meant to address. - Confirm the input is actually a
MASK. Node colors may be a clue, but use the displayed type label as the check. - For a fixed region, provide one mask frame. The documented repeat behavior means you do not need to duplicate it to match the output length.
- For changing position or silhouette, prepare multiple mask frames corresponding to the relevant source interval. Check the sequence’s start, end, and visible subject coverage rather than assuming it spans the whole job.
- Compare the offset with the mask frame count. The documentation says an input is ignored when the offset is greater than or equal to its frame count, with a specific exception for a single-frame
character_mask. Do not assume that a multi-frame mask is repeated when its available frames run out; the documentation does not define that behavior.
The node documentation describes these offset and single-frame rules in its WanAnimateToVideo input and constraint reference. It also identifies length as the number of frames to generate, which is useful when defining the interval to review.
Use a coverage worksheet
The figures below are hypothetical planning data, not a software test or guaranteed frame-selection result. Record what you supplied and what the output review shows; do not infer an undocumented indexing rule from the worksheet.
| Hypothetical task | Mask supplied | Offset and intended interval | Review observation | Decision |
|---|---|---|---|---|
| Locked camera, subject stays within a fixed region | 1 frame | Offset 0; output length 48 | Same region is acceptable at opening, middle, and end | Keep one frame if all reviewed frames meet the brief |
| Subject walks across frame | 60 frames | Offset 24; intended source coverage is 24–59, which is within the 60 supplied frames | Opening mask looks aligned; middle and final coverage still need review | Confirm actual later-frame alignment before accepting the sequence |
| New chunk begins after available mask frames | 40 frames | Offset 40; another 32 output frames requested | Documentation says the multi-frame input is ignored at this offset | Supply a sequence that covers the intended interval or reassess the offset |
| Performer turns and raises an arm | 48 changing frames | Offset 0; output length 48 | Torso aligns, but raised-arm coverage is uncertain | Inspect representative pose changes; do not treat this as automatic correction |
For your own worksheet, replace the hypothetical values with the mask frame count, offset, output length, and interval you intend to cover. The observation column is a review record, not a prediction of how every frame is indexed.
Limits and next steps
A single frame is appropriate only when repeating the same region is acceptable. It cannot encode a changing boundary. A multi-frame mask supplies time-varying regions, but the documentation does not claim that the node generates those masks, tracks the person, or repairs inaccurate outlines. Nor does it specify how a short multi-frame mask behaves after the available frames are exhausted, apart from the offset-at-or-beyond-count ignore condition.
If a moving subject is uncovered, first compare the intended source interval with the supplied sequence and current offset. If the offset is at least the sequence’s frame count, address that documented ignore condition. Otherwise inspect the actual mask frames and output at the beginning, middle, and end; if those masks do not describe the desired regions, prepare a better sequence using your chosen mask-authoring method. Keep the intended length in view: a correct mask at the first frame does not establish coverage throughout the requested output.
The node is marked experimental in the documentation. Its video_frame_offset output provides an updated value for the next chunk, but this guide does not infer that every other input will have identical chunk behavior. See the official node reference for its stated input and output behavior.
Common questions
Should I duplicate one mask frame to match `length`? No, not if a fixed region is intended. The documented single-frame behavior repeats that frame for every output frame.
Does a multi-frame mask automatically follow the character? No such automatic tracking is documented. The sequence itself must represent the changing regions you intend to provide.
What if the offset is beyond my multi-frame mask? The documentation says that input is ignored when the offset is greater than or equal to its frame count. Check the sequence coverage and offset before running; do not assume the node loops or repeats a multi-frame mask.
Sources and next step
This guide interprets public documentation and uses explicitly hypothetical acceptance examples, not application tests. Controls and availability can change; verify the cited manual against your installed version.
For work created in Imagild, carry the input requirements and acceptance checklist into the film workspace. The external software controls discussed above belong to their respective products.