A ten-second check before clicking Generate can save a costly retry. Most avoidable failures come from a handful of things: bad reference files, unclear target mapping, unsupported duration or expectations the selected model cannot meet.
Check the source clip
- Rights: do you have permission to edit and share the footage?
- Duration: is the clip within the model's limits?
- Format: will the provider accept the actual codec and container?
- Faces: can the intended target be tracked across the important shots?
- Sound: will the tool preserve audio, generate new audio or return silence?
If a source clip fails one of these checks, fix that first.
Check the identity references
- Image quality: is each replacement face clear and naturally proportioned?
- Order: does image 1 correspond to the intended target 1, and so on?
- Visible features: do the photos show enough hair, jaw, glasses or facial hair for the edit?
The mapping check is especially important in two-person scenes. Do not assume the model will understand your intention from upload order alone.
Check the generation settings
- Task: is the model configured for a video-editing task, rather than creating a brand-new scene?
- Resolution: are you using a sensible quality level for testing? Low resolution is often enough to verify targeting before a final export.
Finally, write a short prompt that names who changes and who stays the same. More words are not necessarily better words.
Inspect the finished result systematically
Watch it once for the story, once for the target faces, and once for background people and sound. That three-pass review catches more than repeatedly staring at the opening thumbnail.
For source-photo preparation, see our photo reference guide. For multiple actors, follow the two-person mapping workflow.
