Source-photo quality is one of the few inputs you directly control in a focused Hotel Lobby workflow, so it has an outsized effect on the result.
The goal is not to find a perfect studio portrait. The goal is to give the model an unambiguous identity reference with enough visible facial and body information.
Use one clear subject per photo
Group photos create ambiguity because the model must infer which face belongs to the intended performer. Cropping a group photo can help, but a separate single-subject image is usually cleaner.
The same principle applies to pets: keep one main animal clearly visible in each source image.
Show the face clearly
Avoid sunglasses that hide most of the eyes, heavy motion blur, extreme side profiles, hands covering the face, or aggressive beauty filters. Those details can remove identity information the model needs.
A normal phone photo can work well if the face is large enough to see and the lighting is not extremely dark.
Include enough of the body
If the generated motion includes upper-body or full-body movement, a wider source image gives the model more information about clothing, proportions, and pose.
A tight headshot can still work, but the model has to invent more of the body and outfit outside the visible crop.
Make the two performers visually distinct
If two source photos have nearly identical hairstyles, clothing colors, and framing, identity drift can become harder to notice and harder for the model to resolve.
You do not need deliberately different outfits, but clean visual separation between the two references can help.
Do a low-cost test before exporting higher quality
If you are unsure whether a photo pair will work, generate a lower-resolution test first. Once the identities and left/right placement look good, move to a higher export quality for the version you want to keep.
