Ecommerce Model Replace Images

Ecommerce Model Replace Images

When to Use

Use this skill when the user wants to replace or transform the model/person/product relationship in an ecommerce image. Typical requests include:

• Keep the product or outfit and replace the model’s appearance, age range, styling, face direction, or audience type.
• Keep a model pose/composition and put the user’s product onto that model.
• Convert a competitor-style model image into a clean image for the user’s product.
• Keep the scene or pose while changing the person, product, or styling.
• Create different audience versions of the same product, such as commuter, vacation, young, mature,欧美,日韩, or 1688/Taobao/Xiaohongshu style.

Use this for replacement/compositing workflows with at least one reference image. Use ecommerce-ai-model-images when no model reference exists and the goal is to create a new model shot from product photos. Use ecommerce-pose-variant-images when the person/product should stay consistent and only the pose changes.

Inputs

• Required: product or source image. Preserve the product’s true color, print, shape, logo placement, pattern, garment silhouette, fabric drape, accessories, and distinctive details.
• Required: replacement instruction. Name what changes and what stays unchanged, for example 保留罩袍，替换为亚洲轻熟模特, 保留姿势，把商品换成图1, or 保留场景，换成自家包.
• Optional: model reference image, target model image, pose reference, or scene reference. Use references only for the intended role; do not copy competitor marks, watermarks, text, or unrelated products.
• Optional: target platform, target audience, country/region style, scene, output size, output count, and format.
• Optional: face/identity instruction. Do not impersonate celebrities, public figures, or real private persons unless the user provides authorization and a suitable reference.

Before prompting, separate preserve, replace, reference-only, and forbidden. If replacement intent is ambiguous, ask one focused question.

Execution

Use $gpt-image-2 through the bundled edit script. Prefer:

python3 "/Users/yunshi/.codex/skills/gpt-image-2/scripts/edit_image.py" \
  --prompt-file "<prompt.txt>" \
  --image "<primary-product-or-source-image>" \
  --image "<optional-model-or-pose-reference>" \
  --out-dir "<output-dir>" \
  --name "model-replace-01" \
  --output-format "png" \
  --quality "high" \
  --size "1024x1536"

Use the Library path when operating from the canonical oo skill store: /Users/yunshi/Library/Application Support/oo/skills/registry/gpt-image-2/scripts/edit_image.py.

Prompt structure:

Create a marketplace-ready ecommerce model replacement image.
Primary product/source: preserve image 1 accurately, including product shape, color, print placement, garment silhouette, fabric drape, logos, labels, and distinctive details.
Replacement goal: <what changes>.
Must preserve: <product | pose | scene | composition | lighting>.
Reference usage: use images 2+ only for the specified model/pose/scene/style role; do not copy competitor brands, text, watermarks, or unrelated products.
Model requirements: <age range, gender, styling, region aesthetic, body/pose constraints if supplied>.
Composition: natural ecommerce model photo, product clearly visible, realistic anatomy, believable fit/interaction, platform-safe styling.
Avoid: celebrity/private-person impersonation, distorted face/body/hands, changed product pattern/color, fake logos, extra products, QR codes, phone numbers, watermarks, platform badges, unsupported claims.

Result Handling

The script returns JSON. Read local_paths, remote_urls, uploads, and metadata. Save outputs under /Users/yunshi/Downloads/ecommerce-model-replace-images/<short-product-name>-<timestamp>/ and preview the first image when practical.

Failure Handling

• Missing source/product image: ask for it.
• Ambiguous replacement target: ask what should change and what should stay unchanged.
• Identity/face risk: avoid public figures and unauthorized private-person impersonation.
• Product changes too much: retry with stronger preservation wording and product image first.
• Script failure: report the JSON error, stderr progress, and smallest next fix.