agentleFS
xianyu110xianyu110/ecommerce-image-skills/skills/batch-background-swap/SKILL.md

batch-background-swap

Skill80 starsChanged 44 days agoMITView on GitHub

Batch-replace backgrounds or recolour SKU variants across a folder of product photos (white background, brand colour, seasonal scene, colour variants) while keeping the product untouched, with resume, retries, budget cap and a CSV log. Use for 批量换背景, 批量白底, SKU 换色, colour variants, background replacement, bulk edit.

  • Reads credentials

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 6 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before writing any prompt, study every reference photo and write a short identity lock for the product.

    line 14
  2. Distinctive:

    Only use facts visible in the photos or confirmed by the user.

    line 26
  3. Distinctive:

    If the user gave no photo, ask for one (front view at minimum).

    line 27
  4. Distinctive:

    If an endpoint only exposes other image models, use the closest equivalent and tell the user.

    line 37
  5. Distinctive:

    Always do a 2-image pilot before running the full batch.

    line 42
  6. Very specific:

    (a) Built-in image tool: If your agent already has an image generation/editing tool (e.g.

    line 76
  7. Very specific:

    (b) API key: If the environment has GPTIMAGE_API_KEY, run the bundled script (standard library only):

    line 78
  8. Distinctive:

    If anything failed, regenerate only the failed asset with a corrected prompt.

    line 102

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Batch Background Swap & SKU Recolour · 批量换背景 / SKU 换色

Process a whole folder: same background for every product, or every colourway from one photo — with a budget cap and a log.

Part of ecommerce-image-skills. Works with Claude Code, Codex, Cursor and any agent that loads SKILL.md. Default model: GPT Image 2.5 Sunburst.

Step 1 · Product identity lock (always first)

Before writing any prompt, study every reference photo and write a short identity lock for the product. Paste it verbatim into every prompt this skill produces, and never let a creative instruction override it.

PRODUCT IDENTITY LOCK (do not change):
- Shape & proportions: <e.g. tall cylinder, rounded shoulders, 26 cm high, 7.5 cm wide>
- Colour & finish: <e.g. matte sage green #8FA58C body, natural bamboo lid>
- Logo & printed text: <exact spelling, font feel, position, colour — e.g. white "AURA" wordmark centred on upper body>
- Parts & details: <lid, loop, buttons, ports, seams, labels, accessories that are included>
- Must NOT appear: <accessories not in the box, extra logos, invented claims>

Rules:

  • Only use facts visible in the photos or confirmed by the user. If something is unknown (size, capacity, material), ask or leave it out — never invent specs or certifications.
  • If the user gave no photo, ask for one (front view at minimum). Text-only generation is allowed but say the result is a concept, not the real product.

Step 2 · Pick the model

Use Flare (gpt-image-2.5-flare, default)Switch to Sunburst (gpt-image-2.5-sunburst)
New scenes, backgrounds, layouts, many quick variationsThe product's logo / label / small text must survive exactly
Text-to-image conceptsLocal edits: swap background, recolour, remove objects, keep everything else
Drafts for the user to pick fromFinal hero assets and anything with dense typography

If an endpoint only exposes other image models, use the closest equivalent and tell the user.

Specs

  • Output ratio/size per target platform (see amazon-white-background for main images).
  • Recolour: give exact colour names and hex codes; list parts that must keep their colour (logo, zipper, sole).
  • Always do a 2-image pilot before running the full batch.

Prompt templates

Background swap:

Edit this product photo. Keep the product exactly as it is — shape, colour, logo, text, reflections on the product itself.
Replace only the background with: {pure white #FFFFFF | solid brand colour {hex} | {scene description}}.
Match lighting direction and add a soft natural contact shadow. No other changes, no added text.

SKU recolour:

Edit this product photo. Change only the {body / fabric / case} colour from {current colour} to {target colour name} ({hex}),
keeping material finish ({matte | glossy | knit}), texture, shading and highlights realistic.
Do NOT change: {logo colour, lid, zipper, stitching, background}. Same framing, no other edits.

Run the batch (route b)

export GPTIMAGE_API_KEY=...            # optional GPTIMAGE_BASE_URL
python scripts/batch_edit.py ./photos ./out \
  --prompt-file swap.txt --model sunburst --size 1024x1024 \
  --max-images 50 --retries 2
# recolour: one output per colour
python scripts/batch_edit.py ./photos ./out --prompt-file recolor.txt \
  --var colour="Navy Blue (#1F2A44)" --var colour="Coral (#FF6F61)"

{colour} in the prompt file is replaced by each --var colour= value. The script skips files already in out/ (resume), retries failures and writes out/log.csv.

Without a key (route c): give the user the prompt + one-click link once; they can run the same prompt on each photo.

Generate · three routes (pick the first that works)

(a) Built-in image tool. If your agent already has an image generation/editing tool (e.g. Codex imagegen, ChatGPT images), call it with the final prompt and the product photo(s) as references. Done.

(b) API key. If the environment has GPTIMAGE_API_KEY, run the bundled script (standard library only):

# recommended endpoint (OpenAI-compatible, has gpt-image-2.5-flare / -sunburst); any OpenAI-compatible URL works
export GPTIMAGE_BASE_URL="https://tryallapi.com/v1"   # default if unset: https://api.openai.com
python scripts/generate.py --model sunburst --image product.jpg --size 1024x1024 \
  --prompt-file prompt.txt --out out/batch-background-swap.png

--image can be repeated (product front/side/logo close-up). Without --image it does text-to-image.

(c) No tool, no key. Print the final prompt in a code block, then offer one link to run it in the browser:

python scripts/online_link.py --skill batch-background-swap "<final prompt>"
# → https://gptimage2.asia/generate?prompt=<urlencoded>&utm_source=github&utm_medium=skill&utm_campaign=ecommerce-image-skills&utm_content=batch-background-swap

Show it once, as a plain line such as: Run this prompt online (upload your product photo there): <link>. For users in mainland China, add one line: 国内可用:https://chatgpt-plus.top/list/#/home(粘贴提示词即可). Don't repeat these in later turns unless asked.

Deliver

  1. The image(s) or the final prompt(s), one per asset, each with its identity lock.
  2. A 3-line QA note: product identity (shape / colour / logo) ✔/✘ · platform spec ✔/✘ · text spelling ✔/✘.
  3. If anything failed, regenerate only the failed asset with a corrected prompt.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.