Image prompting
- What happens
- Every ad starts as a text-free photoreal plate, generated to order.
- Tool
- Google Nano Banana Pro 2 at 2K, free and unlimited in the Magnific browser, AI-prompt enhancer OFF so the literal prompt survives. Aspect 4:5 for the feed. Everything on this pipeline stays on a free route by design.
- Why
- Photoreal editorial style, warm natural light, shallow depth of field. Subject ethnicity and age are specified explicitly, which is how the set-wide diversity distribution gets built deliberately instead of left to model defaults. Theme is three decisions, not one, and all three come from the same source: what the event actually is plus the promise the copy makes. First the venue, the room itself: identify the setting and its character, then write the prompt to it, so an Italian restaurant becomes a warm rustic trattoria, terracotta and dark beams and olive trees in pots, never a generic interior. Second the character, meaning what the PEOPLE are doing, which is the axis most often forgotten because the backdrop feels like the whole job: are they listening or mingling, watching a presenter or facing each other, is the register a reception, a classroom, a workshop, a celebration, an office. A networking event needs networking behaviour; an educational event needs attending behaviour. Third the sound, the music bed's register, matched to the same world. Crossover is fine and usually right, since real events mix registers; contradiction is not. Where venue and character pull apart, the copy decides, because the copy is the offer. This event proved it: the copy promised what agents would learn from the presenter, so the educational character, a presenter addressing a listening group, matched the offer more literally than agents mingling at the bar did, even though both were true of the same afternoon in the same room. Both shipped, as two separate concepts rather than one blended frame. Character also sets motion difficulty: listening means closed mouths and still heads, the safest case for image-to-video, while a celebration register means open mouths and large motion, the hardest. One plate serves both the static and the motion cut of a concept, and it is designed against the motion constraints from the start, because a plate built that way still makes a fine still while a still-only plate often cannot animate. The lower third stays softly darkened and simple because the headline lives there. And every prompt ends with the same clause: "Absolutely no text, no words, no lettering, no logos, no signage, no readable screens anywhere." The studio overlays its own type; generated text is a defect.
- Gate
- Plate QA: faces, hands, artifacts, and scene-recognition fit (does this frame read as the viewer's life in one glance). Credit rule enforced: free route only, never a paid generation without explicit owner sign-off first.