Images

Generate and edit images from your coding agent with GPT Image 2.5. Describe what you want; the skill writes the prompt, generates and saves the file.

The routemux-images skill lets your agent generate and edit images with GPT Image 2.5 and save them to a file. You describe what you want in plain words — the skill writes the detailed prompt for you.

Not set up yet? Follow the quick start first. The image models need an account that has topped up at least once.

Ask for an image

Draw a minimalist coffee shop logo and save it to ./logo.png

What happens next:

It writes the prompt. Your request becomes a detailed prompt covering the subject, style, composition, lighting, and any text that must appear — in quotes, spelled exactly. The agent shows it to you in one line.
It generates right away. No round of clarifying questions, unless the request is truly ambiguous.
It reports back: the path it saved to, the real pixel size, the model and the list price.

A few more requests that work well:

A wide banner for a hiking app: mountains at dawn, the headline "Trail Ready"

A portrait poster for a jazz night, art deco style, save it to ./jazz-poster.png

A flat vector app icon of a paper plane on a teal background

Edit an image

Keep talking, as you would in a chat:

Make the background blue

The skill edits the last image instead of starting over. You can also point it at an image you already have:

Edit ./photo.png and remove the person in the background

Shape and size

The skill picks the shape from what the image is for. To choose yourself, just say "square", "landscape" or "portrait".

Size	Shape	Good for
`1024x1024`	Square	Icons, logos, avatars
`1536x1024`	Landscape	Banners, slides
`1024x1536`	Portrait	Posters, phone wallpapers
`auto`	Model decides	When the shape does not matter

The size sets the aspect ratio. The image you get back can be larger than the size requested — the skill always reports the real size.

Models

Model	Use it for
GPT Image 2.5 Flare (default)	Almost everything. Fast, and OpenAI's recommended default.
GPT Image 2.5 Sunburst	Precise edits — keep the subject, change one thing. Slower.

The skill can switch to Sunburst for precise edits on its own. You can also ask for it by name: "use Sunburst for this edit".

Pricing

Images are billed per image, not per token, so a longer prompt does not cost more. The current price is on each model's page: Flare · Sunburst.

The skill makes one image per request. If you ask for several options, it makes them one at a time and says so first — each one is billed.

Files

The models return WebP or PNG, and the file is named after its real format. Asking for logo.png may save logo.webp.
The skill never overwrites an existing file. If logo.webp exists, the new image is saved as logo-2.webp, then logo-3.webp, and so on.

Daily limit

Each API key has a daily image allowance. When it is used up you get RMX-LIMIT-4004; it resets at 00:00 UTC. See Troubleshooting for this and other errors.