The outfit edit, kept photoreal
The outfit edit, kept photoreal
Higgsfield Academy
Caption
Every test run so far has had the hero in a plain black outfit — fine while everything else was being tested, wrong for the real ad. Now that the character and kitchen are locked, it's time to dress him for real.
Ten outfits from one photo
Instead of describing clothes by hand, drop the locked character sheet into Claude and ask for a batch of casual outfit ideas written as ready-to-run prompts. Claude comes back with ten distinct, fully detailed looks — no guesswork on fabric, fit, or color needed.
Run all ten through GPT Image 2.0 and compare them side by side. About half come back wearable — the point isn't to find one perfect look on the first pass, it's to generate enough real options that two good ones surface.
Combine the winners into one prompt
Don't settle for a look just because it's "good enough." Here, one option had the better top and the other had the better jeans — so the fix is to combine them: take the shirt from one look, recolor it, and pair it with the jeans from the other, all in a single new prompt, then run that back through GPT Image 2.0.
Don't let the edit soften the face
Every edit pass costs quality. The character sheet was built in Soul Cinema, sharp down to the skin texture — running it through GPT Image 2.0 for the outfit edit softens the whole image. That's the flat, plastic look of AI slop, and it's exactly what a real-looking ad can't afford.
The fix doesn't require a better prompt — it requires a second layer. Pull both versions into any basic photo editor: the original high-detail Soul Cinema shot on top, the edited GPT Image version underneath. Mask out just the outfit on the top layer so the new clothes show through from below.
Face, skin, and background all stay the original Soul Cinema render — only the outfit comes from the edited layer. The edit model never touches anything but the clothes.