If an AI dance video has face distortion, first compare the affected segment with the original footage and the assigned portrait. If AI motion transfer is not working at all, identify whether the problem occurs during upload, analysis, portrait setup, generation, or playback. A distorted output and a failed task need different next steps.
This guide describes MoveFrame AI’s reference-MP4 workflow. For the normal sequence, see the AI dance video tutorial.
Find the stage that needs attention
| Symptom | Check first | Next action |
|---|---|---|
| Upload rejected | File format, nonempty file, size | Export a standard MP4 up to 100 MB; portraits up to 10 MB |
| No scenes yet | Whether analysis is still running | Refresh progress before creating another task |
| Generation cannot start | Missing portrait assignments | Assign portraits or choose Keep original |
| Wrong appearance in a scene | Which portrait is assigned | Correct the assignment before another attempt |
| Face deforms during a turn | Original frame clarity and angle | Try a clearer portrait or simpler reference |
| Task shows failed | Task state and available retry | Use the failed-task retry control |
| One output segment looks poor | Whether the task otherwise completed | Regenerate that segment or keep the original |
| Preview does not play | Whether processing completed | Refresh progress and retry playback |
Face distortion during turns or fast movement
Pause the original video at the affected moment. Check whether the face is blurred, in profile, partly out of frame, or hidden. Then compare that view with the portrait. A front-facing photo may be a poor match for a performance dominated by a strong side angle.
Try one change at a time: use a sharper portrait, select one closer to the reference angle, or replace the difficult source shot. Regenerate the affected segment and compare moving frames. Repeatedly generating the same difficult input does not guarantee a better result.
The reference footage guide provides a pre-upload checklist for angle, lighting, blur, framing, and group scenes.
Hands, microphones, or another person cover the face
Inspect the start and end of the occlusion, when a hand or microphone crosses the facial boundary. If the face is repeatedly covered in the source, consider a cleaner performance or keep that segment unchanged.
For two people in one shot, the current tool has no independent identity target for each performer. Different portraits across segments do not solve overlapping faces inside one frame. Use separate solo segments, or preserve the group shot. This is particularly relevant when adapting a Hotel Lobby two-person video idea.
Clothes or backgrounds change even when the face looks right
Character motion transfer can change more than the face. Review clothing edges, hands, body proportions, and scenery through the entire clip. A good thumbnail does not establish consistency across frames.
Keep scenes that need exact original details, such as a logo or title. For performance scenes, simplify the source and assess whether the new output is usable for your project. There is no guaranteed identity-only or background-lock control in this workflow.
Missing portraits or the wrong character
Open the scene settings and check each generated segment. Select the intended portrait in the library and apply it to the current scene, selected scenes, or all generated scenes as appropriate. Uploading a portrait to the library alone does not mean every segment has received that assignment.
Use Keep original for segments you do not want to generate. Review the missing-assignment count before submitting. If a generated segment uses an unexpected character, verify the saved assignment before trying another render.
Upload failures, failed tasks, and slow progress
Confirm that you are signed in. Use a nonempty standard MP4 no larger than 100 MB; portraits must be JPG, PNG, or WebP up to 10 MB each. Export an actual MP4 from your editor if the source format is uncertain.
If progress appears unchanged, refresh the existing task first. Do not assume a long wait means the task has failed, and avoid starting duplicate jobs while the original is still processing. If it explicitly shows failed, use its retry option. If the service is unavailable, retry later. When contacting support, include the task ID, approximate time, stage, and visible message so the failure can be investigated.
Music or mouth movement does not match your expectation
The export merges the uploaded MP4’s source audio. It does not automatically add a song or create a new vocal performance. Audio merging does not guarantee singing lip-sync. Listen to the source file and the full export before deciding whether the issue is in the input or the assembled result.
For song-specific preparation, the Chuttamalle AI video song tutorial explains which inputs you must supply. Check any separately added music in your editing or publishing app.
Before downloading or trying again
Preview the full video after reviewing individual segments. Check the first frame, scene boundaries, soundtrack, and final frame. If the preview fails, refresh progress and try again; use the completed MP4 download when available.
Return to the video workspace with one specific change to test. Keep the same reference when testing a new portrait, or keep the portrait when testing simpler footage. This makes it easier to identify which input improved the result.
