Choose two portraits the model can read
A clear face gives the generator more useful information than a dramatic background. You do not need matching locations, but both people should be easy to recognize.
One person in each portrait
Use two separate images rather than a busy group photo. Keep the face large enough to see both eyes, the nose and the mouth. Avoid sunglasses, hands across the face, motion blur and heavy beauty filters.
Keep useful detail
Natural light and a sharp image are a good starting point. A moderate head-and-shoulders crop usually preserves more context than a tiny face in a full landscape. The browser resizes supported JPG, PNG and WebP images before submission.
Understand the limits
AI can still alter faces, limbs, clothing and movement. Good source photos reduce ambiguity but cannot guarantee a perfect likeness. Use adult photos with permission, and do not upload someone else’s private images without consent.