Change an outfit in a photo
You have a photo — a portrait you generated, a shot from your camera roll — and you want the same person wearing something else. The outfit-change edit does exactly that: one uploaded image plus one sentence, and the model re-dresses the figure while the face, pose, background, and lighting carry over untouched.
What actually happens when you run it
In the studio you attach the photo and type the instruction — "change her outfit to a black evening gown, keep everything else." The model reads the image and your sentence together, identifies the clothing region, and repaints it. Skin, hair, hands, the background, the direction of the light — none of it moves.
That last part is what makes it useful: the identity survives the edit. The same face looks back at you in the new outfit, at the same angle, in the same scene. You are not generating a new person — you are re-wardrobing the one you already have.
Each run takes around 12–15 seconds and bills per image — cents, not dollars. The live price sits next to the button before you spend anything.
Prompts that work
"Change her outfit to a red cocktail dress, keep the pose and lighting." — a full swap where you name the garment and anchor the scene.
"Swap the tee for a fitted black turtleneck." — single-garment edits hold detail best when you name exactly one piece. The smaller the change you describe, the cleaner it lands.
"Same woman, editorial streetwear — oversized blazer, cargo trousers." — naming the style lane (streetwear, eveningwear, workwear) works better than listing ten items.
"Change the jacket to brown leather, matte finish, slightly worn." — fabric and finish are promptable. Leather, satin, denim, knit — the model renders texture if you name it.
How to write the instruction
Think of it like briefing a wardrobe stylist on set: name the garment, the material, the fit, and say what must not change. "Navy silk slip dress, keep the pose, keep the lighting" gives the model a complete brief in one line.
Three things sharpen every edit: the garment name ("a-line midi skirt", not "a skirt"), a material word (silk, leather, ribbed knit), and an anchor phrase — "keep everything else", "same pose, same light" — which tells the model the rest of the frame is off-limits.
Stuck on the words? Hit prompt assist — it rewrites your rough line into a detailed instruction in place, so you can read exactly what the model will see before you spend a credit.
Iteration tips that save credits
Run the cheap model first. Qwen Image 3 Edit is the daily driver — iterate there until the outfit is right, then re-run the winning instruction on Seedream 5 Pro Edit if you need 2K detail.
Change one variable at a time. If the first result is close but the fabric reads wrong, edit the material word and re-run — don't rewrite the whole prompt. Small diffs in the instruction give small, controlled diffs in the image.
Use the result as the next input. Every render lands in your library — pull the best one back into the tray and keep editing it. Outfit → accessories → backdrop can be three cheap edits on the same figure instead of one heroic prompt.
The first frame is the whole game upstream: a clean, well-lit source photo edits better than a busy one. If your source is noisy, generate a clean portrait first and edit that instead.
The honest limits
Occlusion is the hard case — an outfit hidden behind a crossed arm or deep shadow can smear when replaced. Crop or pick a source where the current clothes are visible.
Hands and jewelry near the hem sometimes get partially repainted. If the fingers matter, say "keep the hands unchanged" explicitly.
Edits that target identifiable real people you don't have rights to are refused at the gate — the studio is for your own photos and generated characters.
Everything lands in your local library and auto-deletes after 48 hours — download what you want to keep.
Read next
Change a pose in a photo →