Troubleshooting
Why your fancy cafe AI video looks wrong
Most “bad” fancy cafe renders are not a model problem. They are an input problem, a geography problem or a timing problem. The five fixes below cover almost every result that makes people say “that is not quite it”.
1. The faces drift or blend
What you see: the person in the middle of the clip looks like a cousin of the person in the photo, or both faces slowly become one.
Fix: replace the weaker photo. Use one person per photo, front-facing, sharp, evenly lit, with no heavy filter. If you uploaded a group photo, crop a copy around each person. Keep the two roles straight: Photo 1 is the person seated first, Photo 2 is the person who comes in later.
2. The seat and the table have swapped
What you see: the cup lands on the wrong disc, or the person sits on the table.
Fix: this is a geography failure, not a likeness failure. Fancy Cafe AI’s built-in prompt locks the left disc as the seat and the middle disc as the table. If you are adapting the prompt in another tool, state that rule in plain language and keep the camera wide enough to show both discs. See the prompt guide.
3. The cup floats, disappears or changes shape
What you see: the cup is in the newcomer’s hand, then on the disc, then gone; or it passes through the metal.
Fix: simplify the frame. One cup, one middle disc, no other tableware. A wider shot gives the model more context for contact shadows. If the first render still drifts, retry with the same photos before changing them — the model has some run-to-run variation.
4. The scene looks flat or plastic
What you see: smooth skin, even light, no texture; the cafe looks like a render instead of a room.
Fix: the input photos may be over-filtered. Replace a beauty-filtered selfie with a natural one. If the photos are fine, the issue is the grade: a warm neutral grade, natural window light and 35mm grain are part of the prompt. A second render often helps.
5. The joke does not read
What you see: a nice cafe video, but no story. The cup lands too early, the reaction is missing, or the clip ends before the climb down.
Fix: check the length and the beats. The product renders 21 seconds with five beats: confidence, entrance, cup, realisation, climb down. If you are writing your own prompt, give each beat a time window and end on the climb down, not on the cup.
When to contact support
If the render fails, the credits return automatically. If the video succeeds but is unusable, contact support@fancycafeai.org with your order or video ID and a short description. The refund policy explains what happens next.
Frequently asked
Should I change both photos at once?
No. Change one variable at a time, starting with the photo that looks least like the person.
Can I rerender the same job?
Use “Make another” with the same two photos. Each render is a new job and costs 4 credits.
Why does the video look different from the sample?
The sample is a storyboard preview. Your render uses your photos and the model’s interpretation, so framing and detail will vary.