APIPod logoAPIPod

one output06 / 14

Edit an image

Changes one thing and protects everything else. Most bad edits come from leaving the second half out.

apipod-edit-imageSKILL.md on GitHub

ChatGPTClaudeCodex

When to use it, and when not to

when to use it

  • The user supplies an image and asks for a modification rather than a new picture.
  • Background, lighting, product placement or style has to change while the subject stays recognisable.
  • The label, logo or embossing on a product must survive the edit.

not for

procedure

5 skills

What the skill does, in order.

  1. 01

    Accept the source in any of three forms

    A public HTTPS URL, a data:image/...;base64,... URI, or a bare base64 payload. Inline data is stored before the request is logged, so the job record references a permanent URL.

  2. 02

    Do not invent a URL requirement

    If the user attached a file with no link, you can take the raw image data — say so and ask how they want to provide it. APIPod does not require a URL.

  3. 03

    Write the prompt as two sentences

    First the change, concretely: "Replace the background with a beach at golden hour." Then what stays: "Keep the bottle's shape, label text and reflections unchanged."

  4. 04

    Read the model's own editing fields

    Some want the strength of the change, some take a mask, some take multiple references. Copy the field names exactly into extras — do not assume them.

  5. 05

    Compare and say what actually changed

    check_job every 3–5 s, then show the result next to the source. If the subject drifted — different label, wrong proportions — say so and propose one corrected prompt.

try it

Send it from here

example request

Here's my product photo: https://example.com/bottle.jpg — use APIPod to replace the background with a beach at golden hour and keep the bottle exactly as it is.

ChatGPTClaudeCodex

The buttons open a new chat with this prompt prefilled. If APIPod is not connected yet, the prompt tells your assistant to stop and help you connect first. In Claude Code or a terminal, copy the prompt instead.

Not set up yet? Wiring guides for Claude, ChatGPT, Codex and more →

call trace

  1. list_models (input_types: image)

    → models that accept a source

  2. get_model_params (editing fields)

    → copied into extras verbatim

  3. generate_image (image_urls[0]: source)

    → img_8d13b0

  4. check_job (every 3–5 s)

    → completed

Illustrative: the tools this skill runs, in order, for the request on the left. Prices appear the way estimate_cost returns them live — this page itself quotes nothing.

what it will not do

The rules it holds itself to.

  • 01

    Very large changes — a new pose, a new viewpoint — are closer to generation than editing. Say that and offer apipod-generate-image instead.

  • 02

    Text inside an image is preserved or replaced well by some models and garbled by others. If exact typography matters, check the model list rather than promising it.

  • 03

    Several images given means the first is the source unless the user says otherwise.

Install once, then just ask.

Connect the MCP server to your client. The skills load with the plugin — there is nothing else to configure.