Ecommerce Pose Variant Images
When to Use
Use this skill when the user wants pose fission or pose variants while keeping the same ecommerce product/model identity. Typical requests include:
- Same clothing on the same or similar model in standing, walking, side, back, seated, turning, hand-on-waist, hood-up, arm-extended, or detail poses.
- Same bag shown as hand-carry, shoulder-carry, crossbody, elbow-carry, table placement, or half-body shot.
- Same accessory or beauty product in multiple wearing/use angles.
- Build a consistent model image set for listing images, detail pages, Xiaohongshu notes, or 1688/Taobao model galleries.
Use this when consistency matters more than replacement. Use ecommerce-ai-model-images for the first model creation and ecommerce-model-replace-images when swapping a person/product in a reference.
- Required: one base model/product image or product image. Preserve identity-relevant traits, product color, print, silhouette, pattern placement, logo, fabric drape, styling, and scene rules.
- Required: pose/action direction or number of variants. If the user says only
姿势裂变, choose a small safe set of ecommerce poses.
- Optional: product-only reference for better fidelity.
- Optional: pose reference images. Use only for pose/action/composition, not identity or brand copying unless instructed.
- Optional: elements to keep consistent: model face, hair, styling, product, background, lighting, camera angle, crop.
- Optional: platform, output size, count, and format.
Do not promise exact identity preservation. Phrase it as same/similar commercial model consistency unless the input/reference and permission are explicit.
Execution
Use $gpt-image-2 through the bundled edit script. Generate variants one call at a time when each pose needs a distinct prompt; use --n only for loose variations of the same prompt.
python3 "/Users/yunshi/.codex/skills/gpt-image-2/scripts/edit_image.py" \
--prompt-file "<prompt.txt>" \
--image "<base-model-or-product-image>" \
--image "<optional-product-or-pose-reference>" \
--out-dir "<output-dir>" \
--name "pose-variant-01" \
--output-format "png" \
--quality "high" \
--size "1024x1536"
Prompt structure:
Create a marketplace-ready ecommerce pose variant image.
Base image: preserve the product/model styling from image 1, especially product color, print placement, silhouette, fabric drape, logos, and distinctive details.
New pose/action: <pose>.
Keep consistent: <model appearance/style/product/background/lighting/camera angle as requested>.
Reference usage: use images 2+ only for product fidelity or pose guidance.
Composition: natural ecommerce model pose, product clearly visible, realistic anatomy/hands, clean lighting, platform-safe crop.
Avoid: changed product pattern/color, hidden product, distorted face/body/hands, changed identity beyond requested, extra products, watermarks, QR codes, unsupported claims.
Result Handling
The script returns JSON. Read local_paths, remote_urls, uploads, and metadata. Save outputs under /Users/yunshi/Downloads/ecommerce-pose-variant-images/<short-product-name>-<timestamp>/ using names like pose-variant-standing.png, pose-variant-side.png, etc.
Failure Handling
- Missing base image: ask for one.
- Missing pose list: use 3 safe ecommerce variants or ask if the category is unclear.
- Product/model drift: retry with stronger consistency and product preservation wording.
- Pose is anatomically difficult or unsafe: simplify to a natural commercial pose.
- Script failure: report the JSON error, stderr progress, and smallest next fix.
---
name: ecommerce-pose-variant-images
description: 'Generate ecommerce pose variant images from a model/product image, product references, and pose instructions. Use when the user asks for pose fission, pose variations, multi-angle model shots, walking/sitting/side/back poses, or consistent product model image sets with GPT Image 2.'
metadata:
icon: "\U0001F57A"
title: Ecommerce Pose Variant Images
companionSkill: gpt-image-2
packageName: '@zjxuyunshi/ecommerce-pose-variant-images'
version: 0.0.1
---
# Ecommerce Pose Variant Images
## When to Use
Use this skill when the user wants pose fission or pose variants while keeping the same ecommerce product/model identity. Typical requests include:
- Same clothing on the same or similar model in standing, walking, side, back, seated, turning, hand-on-waist, hood-up, arm-extended, or detail poses.
- Same bag shown as hand-carry, shoulder-carry, crossbody, elbow-carry, table placement, or half-body shot.
- Same accessory or beauty product in multiple wearing/use angles.
- Build a consistent model image set for listing images, detail pages, Xiaohongshu notes, or 1688/Taobao model galleries.
Use this when consistency matters more than replacement. Use `ecommerce-ai-model-images` for the first model creation and `ecommerce-model-replace-images` when swapping a person/product in a reference.
## Inputs
- Required: one base model/product image or product image. Preserve identity-relevant traits, product color, print, silhouette, pattern placement, logo, fabric drape, styling, and scene rules.
- Required: pose/action direction or number of variants. If the user says only `姿势裂变`, choose a small safe set of ecommerce poses.
- Optional: product-only reference for better fidelity.
- Optional: pose reference images. Use only for pose/action/composition, not identity or brand copying unless instructed.
- Optional: elements to keep consistent: model face, hair, styling, product, background, lighting, camera angle, crop.
- Optional: platform, output size, count, and format.
Do not promise exact identity preservation. Phrase it as same/similar commercial model consistency unless the input/reference and permission are explicit.
## Execution
Use `$gpt-image-2` through the bundled edit script. Generate variants one call at a time when each pose needs a distinct prompt; use `--n` only for loose variations of the same prompt.
```bash
python3 "/Users/yunshi/.codex/skills/gpt-image-2/scripts/edit_image.py" \
--prompt-file "<prompt.txt>" \
--image "<base-model-or-product-image>" \
--image "<optional-product-or-pose-reference>" \
--out-dir "<output-dir>" \
--name "pose-variant-01" \
--output-format "png" \
--quality "high" \
--size "1024x1536"
```
Prompt structure:
```text
Create a marketplace-ready ecommerce pose variant image.
Base image: preserve the product/model styling from image 1, especially product color, print placement, silhouette, fabric drape, logos, and distinctive details.
New pose/action: <pose>.
Keep consistent: <model appearance/style/product/background/lighting/camera angle as requested>.
Reference usage: use images 2+ only for product fidelity or pose guidance.
Composition: natural ecommerce model pose, product clearly visible, realistic anatomy/hands, clean lighting, platform-safe crop.
Avoid: changed product pattern/color, hidden product, distorted face/body/hands, changed identity beyond requested, extra products, watermarks, QR codes, unsupported claims.
```
## Result Handling
The script returns JSON. Read `local_paths`, `remote_urls`, `uploads`, and `metadata`. Save outputs under `/Users/yunshi/Downloads/ecommerce-pose-variant-images/<short-product-name>-<timestamp>/` using names like `pose-variant-standing.png`, `pose-variant-side.png`, etc.
## Failure Handling
- Missing base image: ask for one.
- Missing pose list: use 3 safe ecommerce variants or ask if the category is unclear.
- Product/model drift: retry with stronger consistency and product preservation wording.
- Pose is anatomically difficult or unsafe: simplify to a natural commercial pose.
- Script failure: report the JSON error, stderr progress, and smallest next fix.