Start here
Best photos for a fancy cafe AI video
A fancy cafe AI video is only as good as the two faces it has to keep. The model is not retouching a single portrait; it is rebuilding two people across 21 seconds, from several angles, in a scene with a cup, a wall disc and a moving camera. Clear input photos do most of the work.
The two-photo rule
Use one photo per role. Photo 1 is the person seated first; Photo 2 is the person who comes in later. They do not need to know each other. Do not upload a group photo and hope the model picks the right person. It cannot know which face belongs to which role, and the result is usually a blend of both.
If you only have a group photo, crop a copy around each person before uploading. Keep the crop generous enough to include the head, shoulders and upper body.
The checklist
- One face per photo. No group shots, no photobombers, no mirrored selfies with a second face in the background.
- Facing the camera. A front or three-quarter view. A pure side profile gives the model too little to lock onto.
- Eyes visible. No sunglasses, no hair across the eyes, no mask.
- Even light. Daylight or soft indoor light. Avoid a hard flash that flattens the face or a backlit window that turns it into a silhouette.
- Sharp focus. A slightly soft photo becomes a soft face in every shot.
- No heavy filters. Beauty filters, face-tuning and strong colour grades are copied into the render.
- Upper body included. The model needs the shoulders and clothing, not just a floating head.
Match the pair
The two photos do not need to be from the same day, but they should agree on a few things: similar light direction, similar colour temperature, and a similar crop. A warm sunset selfie next to a cold office photo makes the final grade look uneven, even when both faces are clear.
Keep each person’s outfit simple. A plain shirt or jacket reads better than a busy pattern, and it gives the model fewer details to reinvent between shots.
What to avoid
- Public figures, celebrities or anyone who has not agreed to appear.
- Photos of minors.
- Photos with text, stickers or a heavy watermark over the face.
- Extreme angles, fisheye lenses or a face partly out of frame.
- Two photos of the same person in the two slots — the scene needs two roles.
If the first render drifts
Change one variable at a time. Start with the photo that looks least like the person: replace it with a sharper, front-facing image and keep the other photo unchanged. If both faces are right but the scene feels flat, the problem is usually the scene prompt, not the photos — see Why your fancy cafe AI video looks wrong.
Frequently asked
Can I use a selfie?
Yes. A sharp, front-facing selfie with even light is one of the best inputs.
Can I use a photo with a hat?
A hat is fine if the face is clearly visible. A cap that hides the eyes is not.
Do the two photos need the same background?
No. The background is replaced by the cafe. The faces, hair and clothing are what carry over.