ugc-photo-prompts

UGC-Style Photo Prompts

Core principle

Image models default to polish: symmetric features, smooth retouched skin, studio lighting, clean backgrounds. That polish is the single biggest tell that an image is AI-generated, and it's exactly what kills performance for UGC ads and influencer content — the entire point of UGC is that it doesn't look produced.

Telling the model "make it look realistic" doesn't work, because "realistic" isn't a style the model can target — it's an absence of all the polish defaults, and the model needs that absence written into every slot of the prompt as a specific physical detail. Realism comes from specificity (a named flaw, a named light source, a named texture), not from the word "realistic" itself.

The 9-slot formula

Fill every slot, in this order, for every prompt:

1. Format declaration — camera + shot type + aspect ratio, stated first. This locks the model into "photo," not "illustration." e.g. "Ultra-realistic iPhone front-camera selfie, 4:5 vertical" / "Ultra-photorealistic flash photo, 4:5 vertical."
2. Framing & lens behavior — distance (arm's length / chest-up / full body), angle (slightly above or below eye level), and the lens distortion that comes with it (wide-angle selfie warp, a shoulder cropped out of frame, imperfect centering).
3. Subject — the imperfection stack. Never write "beautiful woman, smiling." Specify hair texture and disorder (damp, windblown, flyaways), skin (tone + pores + one specific flaw — freckles, redness, sweat, under-eye shadow, tan line), and a momentary, specific expression (mid-laugh, looking off to the side, mouth slightly open) instead of a generic one.
4. Wardrobe + accessories, named specifically. Fabric, fit, and at least one small accessory (hoops, a layered necklace, a bag strap). Generic clothing description reads as generic AI output — specificity here is doing real work, not just flavor.
5. Pose as interaction, not posture. Give the subject something to physically do — hold an object, mid-gesture, walking, getting their hair touched by someone off-frame — rather than "posing for camera." Static posing is the second-biggest AI tell after skin polish.
6. Setting with texture and clutter. Name actual materials (brick, stone, fridge glass, car upholstery, a tiled floor) and at least one incidental detail unrelated to the subject (a second person's arm in frame, a parked car, a visible price tag). Staged sets don't have incidental detail; real environments always do.
7. Lighting, named and physical. State the actual light source (direct flash, golden-hour side light, fluorescent overhead, window light) and what it visibly does to skin and shadow — not "good lighting," but what the light is doing.
8. Vibe tag. One short phrase tying the scene together — "everyday building-corridor look," "candid basement-party energy" — this keeps the model from sliding back into editorial polish even after all the specific detail above.
9. Negative space. Say what NOT to do. This is doing as much work as the rest of the prompt — see the standing list below.