APIPod logoAPIPod

one output04 / 14

Generate one image

One picture, or one edit of reference images. Stops at one result when one result is what was asked for.

apipod-generate-imageSKILL.md on GitHub

When to use it, and when not to

when to use it

  • A single-image request with a clear outcome: product shot, illustration, social graphic, style transfer.
  • The user has not named a model and wants you to pick by the job.
  • The picture needs exact text in it — a logo or a headline that must be readable.

not for

procedure

6 skills

What the skill does, in order.

  1. 01

    Pick by the job, not by the leaderboard

    Nano Banana for default, fast and cheap; GPT Image or Recraft when exact typography matters; Seedream for premium product and fashion photoreal; a model with image input when editing.

  2. 02

    Read the schema once

    get_model_params once per model per task. Resolution and quality enums, n, seed and any editing fields come from there — never from memory.

  3. 03

    Settle the brief in one pass

    Subject, style, aspect ratio, on-image text quoted exactly, and whether references are edited or only steer the look. Infer defaults instead of turning this into a questionnaire.

  4. 04

    Quote when price is in question

    estimate_cost only when the user asked about cost or set a budget. Map the planned settings into dimensions as strings, e.g. { n: "2", resolution: "2K" }.

  5. 05

    One call, then poll

    n only when variants were requested — each one costs separately. check_job every 3–5 s until completed or failed, then report cost_usd from the job.

  6. 06

    Name what is off, propose one fix

    Describe the picture, not the request. When the result misses, propose one targeted change and generate again only on a yes.

try it

Send it from here

example request

Use APIPod to create a studio product photo of a matte black water bottle on a wet slate surface, soft top light, 85mm, no text.

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 (category: "image")

    → nano-banana-2 (default pick)

  2. get_model_params (resolution · n · seed)

    → input_schema

  3. generate_image (1:1 · prompt)

    → img_2c91e4

  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

    Aspect ratio comes from where the image will run — 1:1 social, 16:9 thumbnail, 9:16 story — not from a default you invented.

  • 02

    Negatives go at the end of the prompt as "no …". Quoted text stays verbatim and says where it goes.

  • 03

    If the user explicitly named a model that is not listed, say so and offer the closest one. Never swap silently.

Install once, then just ask.

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