Do image models follow a detailed outfit? Twelve 2026 outfits on FLUX.3 klein, MOSS-Image and Qwen-Image 2.1
Our first crowd test dressed everyone in plain t-shirts, which made the pictures look dated. So we wrote twelve specific outfits from autumn/winter 2026 trend reports (fit, fabric, layering and colour for each garment, no brand names; drawn from autumn/winter 2026 reports in British Vogue, British GQ and WWDJAPAN) and asked three models to draw each one as a full-body street-style photo.
What we ran
- The same prompt for every model: "A fictional adult wearing [outfit]. Full-body editorial street style photo on a contemporary city sidewalk. The complete outfit and both shoes are visible. Natural daylight ... All garments are unbranded, fully opaque ..."
- Twelve outfits, one seed (300), 1024×1024, each model at its usual settings, one rented RTX 5070 per model. All 36 pictures succeeded. Krea 2 Turbo failed again (every request timed out); we have not found why.
- Median time per picture: FLUX.3 klein 11.5 s, Qwen-Image 2.1 13.2 s, MOSS-Image SFT 15.8 s.
What we saw
- All three follow the colours and the main garments. Cream knit with brown trousers, the navy jacket over a long untucked shirt, the striped rugby shirt, the olive work coat, the red sweater under a navy coat: each model drew them in the right places and colours.
- "Full-body" is read differently. FLUX.3 klein cut the picture at the chin in almost every outfit, and MOSS-Image often did too, so you see the clothes but not the person. Qwen-Image 2.1 kept the whole person, face included, in most pictures.
- Shape words are the weak point. None of the three drew a clear peplum (a flared waist) on the "waist-shaped peplum blazer"; they drew a normal blazer. "Barrel-leg trousers" came out as ordinary wide trousers.
- Details split the models. For the "zip-through cardigan", MOSS-Image and Qwen-Image drew a zip and FLUX.3 drew buttons. For "a large-check jacket over a smaller-check skirt", FLUX.3 got both check sizes; the other two changed the jacket fabric.
- One collage. For the all-black outfit MOSS-Image drew three people side by side instead of one, as it did in our crowd test.
How much to trust this
- One seed per outfit, so these are observations, not rates.
- We judged the garments by eye from the pictures; there is no automatic score.
- The outfits are our own combinations of reported trends; whether they look current is a matter of taste.