How do you keep two characters consistent in a photo-to-video workflow?

[hotellobbyai26]

When turning two reference photos into a short AI video, which input qualities have the biggest effect on identity consistency?

My current checklist is to match the approximate camera angle, keep each face unobstructed, use similar subject scale, and leave enough space around the shoulders for vertical reframing. I also try to avoid heavily filtered portraits because extra texture can become unstable once the subject starts moving. If the two references have very different lighting or perspective, even a good motion template has to solve too many visual conflicts at once.

Duration seems important too. A 15-second template is often enough for one entrance, synchronized movement, and a closing pose. A 30-second version allows more narrative structure, but it creates more opportunities for facial details or clothing to drift. I therefore start with the shortest version that communicates the idea, review it once for pacing, then again for identity, hands, body scale, and the first and last frames.

For a focused preset workflow, hotel lobby ai video turns two photos into a duo performance with preset motion, camera timing, framing, and generated audio. That makes it easier to isolate the effect of the source images instead of changing choreography and references at the same time.

When a result needs improvement, I change only one variable: replace one photo, reverse the subject order, or switch duration. What other checks have helped you keep two characters visually stable across a complete clip?