GPT Image 2 skill for Claude Code: how to use it

What GPT Image 2 is, in one paragraph

GPT Image 2 is OpenAI's image model for generating a picture from a description or editing an existing one with a reference. Claude Code cannot call it on its own. The Actionway Skill — a set of instructions Claude Code reads so it knows which services exist, what they cost, and how to run them — adds GPT Image 2 to the session along with the other image models, so you ask in plain words and get a file back.

A real run: the product photo

你說

Use GPT Image 2: product photo of a single white low-top sneaker with a gum sole on a warm beige seamless backdrop, soft studio light, slight three-quarter angle, no text

AGENT 回覆

GPT Image 2 is in the catalog. Its price, with two neighbours for comparison:

• GPT Image 2$0.15 per rungeneration and precise edit instructions
• Seedream 5 LTE$0.10 per runfast drafts
• Banana 2$0.20 per rungeneration and reference-based edits

你選定: GPT Image 2

• Sends the request to GPT Image 2 — it starts as a background job
• Waits for it; this one finished in 75 seconds
• Saves the image into an outputs/ folder in your project (that is where every generated file lands) and hands you the file

結果

1 image · GPT Image 2 · $0.15 per run · charged from the wallet after completion

That is a real Claude Code session, unedited, with the model named up front. The agent still showed the neighbours and their prices — it always does — but because you named the model, the run went straight ahead. The photo did what a product shot has to do: one object, clean backdrop, no invented lettering, and enough detail in the leather and the sole to survive being edited later. That is why it became the reference for the three ad versions and the five-second product clip in this series. The GPT Image 2 model page is the reference for its price and inputs, and for where it sits in the image generation capability.

When to pick it

Pick GPT Image 2 when the brief has precise instructions — a specific angle, a clean backdrop, text placement, an edit you can describe in words — and when one strong final image matters more than the price of an attempt. Pick Seedream when you are still exploring and want cheap drafts, and the Banana models when the job starts from a reference photo. The agent lists all of them with prices and does not choose for you; say "you pick" and it takes the best-quality one.

Two things people expect to need and do not: an OpenAI API key — the run is billed to your Actionway wallet, so there is no key to create or leak into a repo — and a second install per model. One Skill covers the whole image row, so switching models is a word in your request. GPT Image 2 also edits from a reference image at the same price per image.

What to put in a product-photo brief

The prompt in the run above is a good template because every phrase does a job. "Single" and "one sneaker" stop the model from adding a pair or a second angle. "Seamless backdrop" and a named colour give you a background that is easy to cut out or replace later. "Soft studio light" avoids hard shadows that fight with whatever scene the ad ends up in. "Slight three-quarter angle" is the angle catalogues use because it shows the toe, the side, and the sole at once. And "no text" matters more than it looks: image models like to add lettering to shoes, boxes, and walls, and removing it afterwards costs another run.

Leave out what you do not need. Every extra adjective is one more thing the model may satisfy at the expense of the one you cared about, so a brief of one subject, one background, one light, and one angle usually beats a paragraph. If you want a second version, change one phrase at a time — the colour of the backdrop, the angle — so you can tell what each change did; at $0.15 per run an image, a set of four variations is still pocket change.

Set it up once

npm install --global @actionway/cli@latest
actionway init
mkdir -p "$HOME/.claude/skills"
cp -R "$HOME/.actionway/skill/actionway" "$HOME/.claude/skills/actionway"

You need Node.js 20 or newer and an Actionway account; after sign-in the agent tells you your wallet balance and, if it is empty, links you straight to top-up. Prefer a plugin or a Remote MCP server to the command line? Both end at the same catalog and the same wallet — the get started page lists the steps per client side by side.