ugc-photo-prompts

Use the 9-slot prompt formula: declare camera format first, then specify framing, subject imperfections like pores and flyaway hairs, named wardrobe, an interactive pose, a cluttered setting, a named physical light source, a vibe tag, and a negative prompt banning studio polish and retouched skin.

Call get_timeline to match the canvas aspect ratio, check get_media for an existing product reference image, pick nano-banana-pro for realism or gpt-image-2 for product clarity, then run generate_image and verify completion before placing the asset on the timeline.

nano-banana-pro holds skin micro-texture and follows lighting instructions more faithfully at higher resolution, making it better for UGC realism. gpt-image-2 frames the product more prominently, suiting hero product shots. When it matters, generate one of each and compare.

Reference images passed via referenceMediaRefs anchor the subject's identity like shape, color, and material, but not style or lighting. Describe the scene fully in the prompt, refer to the reference generically, and use clean evenly-lit catalog shots for the best results.

The word realistic does not work because models default to polish. Append the universal realism add-on mentioning visible pores, flyaway hairs, motion blur, phone HDR, and compression artifacts, and verify each prompt slot against the actual output since framing instructions are the least reliable.