fashion-model-replacement
Replace the person or mannequin in an already-dressed fashion image while preserving the complete outfit, garment construction, fit cues, branding, pose, crop, and requested scene. Use when a user ask
SKILL.mdskills/fashion-model-replacement/SKILL.md
GitHubFashion Model Replacement
Produce one dressed-model image. Define the Skill by the replacement job and preservation contract, not by aimodel, a model library, or any single engine.
Catalog
- Display name: Fashion Model Replacement
- Categories: Fashion, Commercial Production
- Status: Ready
- Route label: WeShop AI Model or GPT Image 2
- Tone: coral
- Short description: Replace a fashion model while keeping the outfit intact.
- Similar skills: change-pose, virtual-try-on
What this skill does
- Replaces a person or mannequin already wearing the product.
- Generates a suitable adult model from a concise description when no target-model image is supplied.
- Routes between the currently verified WeShop AI Model contract, newer verified WeShop replacement routes, and GPT Image 2 according to inputs.
- Verifies outfit, anatomy, identity/reference, pose, crop, and scene continuity.
How to use
Provide one already-dressed fashion image, then describe the new model or attach a supplied target-model reference.
Describe a new model
Replace the model with an adult Black woman with short natural curls and keep the complete outfit, pose, and studio unchanged.
Use a model reference
Replace the person in this fashion image with the supplied model reference while preserving the outfit and composition.
Replace a mannequin
Turn this dressed mannequin image into a natural adult fashion-model photo without changing the clothing.
Workflow
1. Confirm the replacement job
Require an already-dressed person or mannequin source. Compile:
source: dressed-person | dressed-mannequin
target_model: supplied-reference | text-description | minimally-generated-adult
replace: whole model | mannequin-to-human
outfit_locks: garments | fit cues | accessories | footwear | branding
image_locks: pose | crop | camera | background | lighting | composition
delivery: count | ratio | resolution
Do not route a separate garment plus person here; use Virtual Try-On. Do not route pose-only work, outfit invention, or a general campaign reshoot here.
2. Build the model brief
Read references/model-brief-and-acceptance.md. When no target reference is supplied:
- use the user's explicit adult model description;
- ask one short question only when target audience or model presentation materially changes the commercial result;
- otherwise generate a neutral adult commercial fashion model with natural features and expression;
- do not browse or expose WeShop model/location libraries without a preview surface;
- pass an existing user-supplied
fashionModelIdonly when the user explicitly provides it.
3. Select the route and mask before spending a run
- Supplied target-model reference: prefer
gpt-imagev1.0 / GPT Image 2 because it accepts the dressed source and target-person reference together. Default toquality: "medium",imageSize: "2K", requested/auto ratio, andbatchCount: 1. - Text-described identity/head replacement with background locked: use the best currently documented WeShop AI Model route. The verified route is
aimodelv1.0 withgeneratedContent: "freeCreation",maskType: "autoHumanSegment",pose: "originalImagePose", nolocationId, and a concise model description. This preserves body and background while replacing face/head. - Whole-model or body-type replacement: prefer GPT Image 2 or a verified
customMaskUrlworkflow. Do not useautoHumanSegmentwhen the requested body itself must change. - Whole model may be regenerated and background is released:
autoApparelSegmentmay be used to lock only the complete outfit. Never select it when the background must stay unchanged. - Exact target appearance: prefer GPT Image 2 with a supplied target-model reference. Without a target reference, treat fine hair and facial language as guidance rather than an identity guarantee.
- Mannequin-to-human or difficult free-form replacement: choose between the current WeShop route and GPT Image 2 based on garment segmentation versus multi-reference/prompt control.
- Swap Model & BG /
changemodelbg: use only after its full execution contract is available and verified; do not infer parameters from the marketing page.
Read references/api-parameters.md before constructing an aimodel request. Never use a mask whose protected region conflicts with the user's invariants.
Do not step through WeShop 1.0, WeShop 2.0, and GPT Image 2 as a quality ladder. Choose once from the inputs. Record the reason for every fallback.
4. Compile one concise Prompt
Text-described WeShop route:
Replace the current model identity with [short adult model brief]. Preserve the original body, complete outfit, pose, hands, crop, background, lighting, shadows, camera, and composition unchanged.
GPT Image 2 with reference:
Replace the person in Image 1 with the adult person from Image 2. Preserve Image 1's complete outfit, [recognition anchors], fit cues, pose, crop, background, lighting, and composition. Integrate the new head and body naturally. One person only; no outfit redesign, extra accessory, text, duplicate, or collage.
Use negative Prompt fields only when the selected route supports them. Do not submit demographic checklists, beauty scoring, or generic fashion adjectives that do not affect the requested result.
5. Execute and accept
Upload real originals, submit one result, poll to terminal state, and inspect the image. Keep the dressed source and any target-person reference as separate Original assets.
Apply the six checks in references/model-brief-and-acceptance.md. API success is not acceptance.
6. Retry one failed property
- outfit drift: reduce model/scene language and reassert only lost garment anchors;
- incorrect target appearance: restate two or three observable model attributes or strengthen the supplied reference role;
- anatomy or neck seam: target that junction without changing outfit or scene;
- scene/pose drift: reassert the original pose, crop, background, lighting, and composition;
- target-reference mismatch on WeShop route: switch to GPT Image 2 with both originals;
- segmentation failure on GPT Image 2: switch to the verified WeShop apparel-mask route.
Never resend an identical request. Record route, model version, parameters, Prompt, execution ID, terminal status, error/refund state, failed acceptance item, and fallback reason.
User-facing output
- Media type: One fashion model replacement image plus a compact source and acceptance record
- Default quantity: 1 separate image
- Content per image: One replacement model wearing the preserved source outfit
- Default layout: Preserve source pose, crop, camera, scene, lighting, and composition
- Video output: No
- Model policy: Text-described replacement uses the best verified WeShop AI Model route; supplied model references prefer GPT Image 2 medium/2K
- Downstream use: Pose variants, fashion photoshoots, lookbooks, product pages, posters, banners, and collages