The trend
The fancy cafe AI prompt, shot by shot
A working fancy cafe prompt is not a list of adjectives. It is a floor plan plus a beat sheet. The model has to understand which disc is a seat, which disc is a table, who belongs to which photo, and when the cup lands. Everything else is style.
The six parts of the prompt
- Cast lock. Name each photo and what it is: Photo 1 is the person seated first; Photo 2 is the person who comes in later.
- Geography. The left wall disc is the seat; the middle disc is the table; the concrete ledge is the bench.
- Beats. Confidence, entrance, cup, realisation, climb down — each with a time window.
- Camera. One locked-off medium-wide shot with a slow push-in, so the two discs stay in frame.
- Audio. Cafe room tone, a ceramic cup on metal, footsteps, a bench scrape; no dialogue.
- Negatives. No extra people, no identity swap, no floating cup, no text, no cartoon style.
Why the geography matters more than the style
Most failed attempts describe “a fancy cafe” and “an awkward moment” without saying where anything is. The model then invents a table, a chair or a third person, and the joke disappears. The prompt has to state the spatial rule in plain language: left disc = seat, middle disc = table.
Once the geography is fixed, the camera can stay wide enough to show both discs. A close-up loses the relationship between the seat and the table, which is the whole point.
The beat sheet
- Beat 1: the seated visitor on the left disc, sipping coffee with confidence.
- Beat 2: the person who comes in later enters from the right and sits on the ledge beside the middle disc.
- Beat 3: the newcomer puts a cup on the middle disc; the cup lands with a ceramic click.
- Beat 4: the first person looks down at the disc, then at the newcomer; a blink, a freeze, no cartoon reaction.
- Beat 5: the first person climbs down, sits beside the newcomer and takes a tight-lipped sip.
A condensed prompt you can learn from
The production prompt is longer, but the structure is this:
Photorealistic 21-second cafe scene, vertical 9:16. Photo 1 is the person seated first; Photo 2 is the person who comes in later. Minimalist concrete cafe: the left wall disc is a seat, the middle disc is a table, a low concrete bench runs along the wall. The first person sits on the left disc with coffee. The newcomer enters, sits on the bench and places a cup on the middle disc. The first person notices, freezes, climbs down and sits beside the newcomer. Locked-off medium-wide shot, slow push-in, natural window light. Produce no audio of your own: render silent. No dialogue, no extra people, no identity swap, no text.
That is enough to show the model the joke. The product adds the negative list, the identity lock and the timing.
Negatives that matter
- No duplicate faces or blended identities.
- No extra people.
- No floating cups, no changing furniture.
- No text, captions, logos or watermarks.
- No cartoon, anime or beauty-filter look.
- No jump cuts that break the geography.
Frequently asked
Do I need to paste a prompt into Fancy Cafe AI?
No. The prompt is part of the product. You choose two photos.
Can I use this prompt in another model?
You can adapt the structure, but identity locking and timing differ between providers. Test with a short clip first.
Why does the cup keep moving?
The model is not a physics engine. A wider shot and a clear “cup on the middle disc” instruction reduce drift; retrying with the same photos is usually faster than rewriting the scene.