gpt-image Skill (inference.sh) for Claude Code: What It Does and How to Install It

Lets Claude generate and edit images with OpenAI GPT-Image-2.5 and GPT-Image-2 through the inference.sh belt CLI.

Installs: 13.2K; on skills.sh
Stars: 17; on GitHub
Updated: Aug 2026; last repo commit

This skill makes Claude call openai apps with belt app run. It lists three apps: GPT-Image-2.5 Flare (the default), GPT-Image-2.5 Sunburst (for precise edits) and GPT-Image-2. It supports text-to-image, editing with reference images, mask-based inpainting, transparent backgrounds and up to 10 images per run, in PNG, JPEG or WebP.

It is a third-party skill, not published by OpenAI. It needs belt installed and signed in, and each run uses inference.sh credits.

Good for

Designers and marketers who want product mockups and visuals from a prompt
Anyone editing a photo, filling a masked region or making a transparent PNG

Install gpt-image

Claude Code

Copy this prompt.

Install the gpt-image skill in Claude Code. Run this command:

npx skills add https://github.com/inference-sh/skills --skill gpt-image

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

Pick one route for the skill, then install the belt CLI with one of the belt routes and run belt login. Every command in SKILL.md is a belt app run call, and its allowed-tools line is Bash(belt *), so the skill does nothing without belt. The folder holds 1 file, SKILL.md (repo tree). It runs belt in a shell, so there is no Claude app upload route.

Built from the package data we hold. Not yet tested by us.

What's in the skill

When it loads

Generate and edit images with OpenAI GPT-Image-2 via inference.sh CLI. Models: GPT-Image-2. Capabilities: text-to-image, image editing, inpainting, mask-based editing, multi-image reference, batch generation. Use for: product mockups, marketing visuals, image editing, concept art, inpainting, photo manipulation. Triggers: gpt image, gpt-image-2, openai image, chatgpt image, dall-e, dalle, openai image generation, gpt image edit, gpt inpainting, openai dall-e, gpt 4o image

Sections in SKILL.md

Quick Start
Capabilities
Examples
Text-to-Image
Multiple Images
Image Editing with Reference
Multi-Image Reference
Inpainting with Mask
Custom Resolution
Fast Drafts
Pricing
Parameters
Related Skills
Documentation

Key facts

Author: inference.sh (owner of the repo; plugin.json lists inference.sh as author)
License: MIT per plugin.json and the README badge; the repo has no LICENSE file
Installs on skills.sh (101-skills/superpowers listing): 13,169 (Nov 2, 2026)
GitHub stars: 797 for the whole repo (Nov 2, 2026)
Last repo commit: Oct 15, 2026
Audits on skills.sh (101-skills/superpowers listing): Gen Agent Trust Hub: pass, Socket: warn, Snyk: pass
What it includes: 1 file: SKILL.md (from the repo tree)
Needs: The inference.sh belt CLI, signed in with belt login
Allowed tools: Bash(belt *)

What it includes

belt app run openai/gpt-image-2-5-flare with a prompt and quality
quality can be low, medium, high, xhigh or max; xhigh and max are 2.5 only.
Try: “Generate a product photo of white sneakers, studio lighting, high quality”
Runs code · Text-to-image
Flare is the default, Sunburst suits precise edits, GPT-Image-2 is the older app
All three apps share the same input schema.
Choose a model
Set n from 1 to 10 to get several images in one run
Example: four logo options for a coffee shop.
Runs code · Multiple images
Pass images with a prompt such as "change the background"
The image inputs in the examples are URLs.
Runs code · Edit with a reference
openai/gpt-image-2-5-sunburst changes only what you ask for
SKILL.md says it is built to preserve the rest of the image.
Runs code · Precise edit with Sunburst
Combine two or more reference images in one scene
Pass several URLs in images.
Runs code · Multi-image reference
Pass images plus a mask to change one region
The mask is an image URL.
Runs code · Inpainting with a mask
background: transparent with png or webp output
SKILL.md says to prompt for an isolated subject.
Runs code · Transparent background
width and height from 256 to 3840 in steps of 16, ratio up to 3:1
Larger sizes cost more.
Runs code · Custom resolution
quality low for quick concept sketches
The cheapest tier in the price table.
Runs code · Fast drafts
belt app get openai/gpt-image-2-5-flare shows full pricing
Run this before a large batch.
Network · Check pricing

From the official docs

Permissions

Medium risk

This skill makes your agent run belt app run commands that send your prompts and files to inference.sh and use credits on your account. Check the price of an app with belt app pricing before you start a large batch.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

Each run uses inference.sh credits
SKILL.md lists approximate prices per 1024x1024 image, from $0.006 (low) up to $0.21 (max on GPT-Image-2.5, high on GPT-Image-2), and says larger resolutions cost more and edits add about $0.008 per reference image. The belt README shows belt balance to check credits and belt app pricing to see the cost formula before a run. github.com
Your prompts and files go to inference.sh
Every belt app run sends its input to an inference.sh app. The belt README says local file paths in an input are uploaded by the CLI. github.com
You must install the belt CLI first
The belt README gives a download-and-review install script, Homebrew, npm and Scoop. cli-install.md in the repo says the installer downloads a binary and checks its SHA-256 checksum. The belt skill named in SKILL.md can also run brew install and npm install -g. github.com

Security advisories

No published security advisories as of Nov 2, 2026.

Technical details

Turning on read-only	SKILL.md describes no read-only mode.
Talks to	inference.sh (belt app run calls). cli.inference.sh and dist.inference.sh (belt installer)
Sign-in and keys	Run belt login to sign in to inference.sh. SKILL.md does not say where belt keeps the login.
Scopes	Runs belt app run commands in your shell (allowed-tools: Bash(belt *)) Uploads the inputs you give it, including local files, to inference.sh

Questions

What does the gpt-image skill do?

It makes Claude generate and edit images with OpenAI GPT-Image-2.5 and GPT-Image-2 through the inference.sh belt CLI. It supports text-to-image, editing with reference images, mask-based inpainting, transparent backgrounds, batches of up to 10 and sizes from 256 to 3840 pixels.

How do I install it?

Use the install box at the top of this page for the skill. You also need the belt CLI from inference.sh, installed with one of the belt routes in the box, and a belt login. There is no Claude app upload route because the skill runs belt in a shell.

Does it need an account or API key?

It needs an inference.sh sign-in through belt login. SKILL.md names no separate API key. For a 1024x1024 image it lists GPT-Image-2.5 at $0.006 (low), $0.013 (medium), $0.053 (high), $0.094 (xhigh) and $0.21 (max), and GPT-Image-2 at $0.006, $0.053 and $0.21 for low, medium and high. Larger sizes cost more, and edits add about $0.008 per reference image.

Is it safe?

It runs belt app run commands through your shell. Each run sends your prompt and any files or URLs you give it to inference.sh and uses credits. The belt CLI has to be installed first, so read its install steps before you run them.

Is this an OpenAI skill?

No. It is not published by OpenAI. The repo owner is inference-sh, and its plugin.json names inference.sh as author. The skill calls openai/gpt-image-2-5-flare, openai/gpt-image-2-5-sunburst and openai/gpt-image-2 on inference.sh.

Who publishes this skill?

inference.sh, in its own repo github.com/inference-sh/skills. The repo's plugin.json lists inference.sh as author and https://inference.sh as homepage.

Is there another copy of this skill?

skills.sh also lists this skill under 101-skills/superpowers, a repo whose plugin.json names inference.sh as author and inference-sh/skills as the repository. This page links to inference-sh/skills, the repo the plugin.json points to.

How much does an image cost?

SKILL.md gives approximate prices per 1024x1024 image. GPT-Image-2.5 costs $0.006 (low), $0.013 (medium), $0.053 (high), $0.094 (xhigh) and $0.21 (max), the same for Flare and Sunburst. GPT-Image-2 costs $0.006, $0.053 and $0.21 for low, medium and high. Larger sizes cost more, edits add about $0.008 per reference image, and belt app get openai/gpt-image-2-5-flare shows full pricing.

Which other skills does it point to?

Its Related Skills section lists infsh-cli, ai-image-generation, flux-image and p-image from inference-sh/skills. They are optional.

Information

updated Nov 2, 2026

Repository: inference-sh/skills
Folder: tools/image/gpt-image

inference-sh/skills skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 6

gpt-image SKILL.md · github.com
belt CLI README · github.com
cli-install.md · github.com
README · github.com
plugin.json · github.com
belt skill SKILL.md · github.com