GPT Image 2 Skill (RunComfy) for Claude Code: What It Does and How to Install It

Makes Claude generate and edit images with OpenAI GPT Image 2 through RunComfy's CLI, with tips for text-heavy images.

Installs: 87.2K; on skills.sh
Stars: 61; on GitHub
Updated: May 2026; last repo commit

This skill makes Claude generate and edit images with OpenAI's GPT Image 2 model, hosted on RunComfy's Model API. Claude builds a runcomfy run command for text-to-image or for an edit with up to 10 reference images, waits for the job and saves the files to a folder. It also carries prompt advice for embedded text, logos and edits that must keep part of the image unchanged.

It is a RunComfy skill, not an OpenAI product. SKILL.md says no OpenAI key is needed, but you need the RunComfy CLI installed and a RunComfy account. Your prompt and any image URLs go to RunComfy. Its folder holds only SKILL.md.

Good for

Designers and marketers who need product shots, ads or posters with readable text
Developers who want Claude to produce UI mockups or localized banners
Anyone who already has a RunComfy account and the runcomfy CLI

Install gpt-image-2

Claude Code

Copy this prompt.

Install the gpt-image-2 skill in Claude Code. Run this command:

npx skills add https://github.com/runcomfy-com/skills --skill gpt-image-2

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 skill route, then install the RunComfy CLI. The first command uses runcomfy-com/skills, RunComfy's own GitHub organisation, which holds the same SKILL.md as the prime-skills copy. The skill only calls runcomfy run, so it does nothing without the CLI and a RunComfy account. runcomfy login opens a browser sign-in. In CI or containers, set RUNCOMFY_TOKEN instead. All three skill commands install a skill named gpt-image-2, from three repos: runcomfy-com/skills, prime-skills/runcomfy-agent-skills (the skills.sh listing our install count comes from) and agentspace-so/runcomfy-skills (the command SKILL.md names). The repo README names agentspace-so/runcomfy-agent-skills, which GitHub now redirects. The skill runs a local CLI, so this page offers no Claude app upload route.

From its listing on skills.sh. Source.

What's in the skill

When it loads

Generate and edit images with OpenAI GPT Image 2 (ChatGPT Images 2.0) on RunComfy. Documents GPT Image 2's strengths (embedded text, logos, multilingual typography, instruction precision), its 3 fixed sizes, edit-with-preservation language, and when to route to a sibling (Flux 2 / Nano Banana Pro / Seedream) instead. Calls `runcomfy run openai/gpt-image-2/text-to-image` or `/edit` through the local RunComfy CLI. Triggers on "gpt image 2", "gpt-image-2", "ChatGPT Images 2", "image 2", or any explicit ask to generate or edit with this model.

Sections in SKILL.md

When to pick this model (vs siblings)
Prerequisites
Endpoints + input schema
openai/gpt-image-2/text-to-image
openai/gpt-image-2/edit
How to invoke
Prompting — what actually works
Where it shines
Sample prompts (verified to produce strong results)
Limitations
Exit codes
How it works
What this skill is not
Security & Privacy

Key facts

Author: RunComfy (LICENSE copyright "RunComfy / agentspace-so"). The identical SKILL.md is in runcomfy-com/skills, RunComfy's own GitHub organisation. We could not confirm who runs the prime-skills organisation that holds the crawled copy.
License: MIT (LICENSE file and SKILL.md frontmatter)
Installs on skills.sh: 87,214 (Nov 2, 2026)
GitHub stars: 61 for the whole prime-skills/runcomfy-agent-skills repo (Nov 2, 2026)
Last repo commit: May 15, 2026
Audits on skills.sh: Gen Agent Trust Hub: pass, Socket: pass, Snyk: pass
What it includes: 1 file: SKILL.md (no scripts)
Needs: RunComfy CLI (npm i -g @runcomfy/cli) and a RunComfy account
Part of: The runcomfy-agent-skills repo: 30 skill folders in the tree, one per RunComfy model or task
Same skill in RunComfy's repo: runcomfy-com/skills has gpt-image-2 with an identical SKILL.md (compared byte for byte). Its skills.sh listing shows 12 installs (Nov 2, 2026), against 87,214 for the prime-skills listing.

What it includes

runcomfy run openai/gpt-image-2/text-to-image with a prompt and a size
Size is 1024_1024 (default), 1024_1536 or 1536_1024. SKILL.md says only these three exist.
Try: “Generate a product shot of a ceramic water bottle with the word "AQUA+" on the label”
Runs code · Text-to-image
runcomfy run openai/gpt-image-2/edit with an instruction and up to 10 image URLs
Reference images must be publicly fetchable HTTPS URLs. Size defaults to auto, which keeps the input ratio.
Runs code · Image edit
Numbers the images in the prompt, such as subject from image 1 and room from image 2
The first image is the primary one and the rest give auxiliary cues.
Network · Multi-reference edits
--no-wait prints a request_id so a job can be checked later
Runs code · Run without waiting
The CLI polls every 2 seconds, then saves outputs to --output-dir
It downloads only from hosts ending in .runcomfy.net or .runcomfy.com, and Ctrl-C sends a cancel request.
Write · Download results
Quote embedded text, keep it short, change one attribute per round, state what to keep
Also: use compositional cues, avoid conflicting instructions, and pick one or two style anchors at most.
Read · Prompt guidance
Points to Flux 2, Nano Banana Pro or Seedream 5 for stylized, portrait or cinematic work
If you explicitly ask for GPT Image 2, SKILL.md says to use it regardless.
Read · Model routing
Explains CLI codes: 64 args, 65 input, 69 upstream, 75 retry, 77 not signed in
Read · Exit codes

From the official docs

Permissions

Medium risk

What it can do

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

Watch out for

Jobs run on your RunComfy account
Each run uses RunComfy's hosted GPUs, and SKILL.md mentions billing for GPU time. It states no prices, so set a spending limit in your RunComfy account if you can. github.com
It needs a globally installed npm CLI
The prerequisite is npm i -g @runcomfy/cli, which installs third-party code on your machine. Install it from the command in SKILL.md and keep it updated. github.com
Your prompts and image URLs go to RunComfy
SKILL.md says image URLs are fetched by RunComfy's servers, not by your machine, and calls image-based prompt injection a known risk. Only pass images you trust. github.com
Security notes in SKILL.md
It says the prompt is sent as JSON without shell expansion, that downloads from other hosts are listed but not fetched, that any single download over 2 GiB is aborted, and that there is no telemetry. github.com

Security advisories

No published security advisories as of Nov 2, 2026.

Technical details

Turning on read-only	SKILL.md defines no read-only mode.
Talks to	model-api.runcomfy.net (job submission). .runcomfy.net and .runcomfy.com (downloads of generated files)
Sign-in and keys	SKILL.md says runcomfy login writes the token to ~/.config/runcomfy/token.json with mode 0600, and that RUNCOMFY_TOKEN bypasses the file.
Scopes	Runs the local runcomfy CLI Sends your prompts and image URLs to RunComfy's Model API Saves generated files into the output folder you give Uses your RunComfy account, or RUNCOMFY_TOKEN, for billing and sign-in

Questions

What does the gpt-image-2 skill do?

It makes Claude generate and edit images with OpenAI GPT Image 2 by calling the RunComfy CLI. It adds prompt advice for text, logos and edits that must preserve parts of the image.

Do I need an OpenAI key?

SKILL.md says no OpenAI key is needed, because the model is hosted on the RunComfy Model API. You need the RunComfy CLI and a RunComfy account instead.

How do I install it?

Use the install box at the top of this page. Add the skill with one of the skill commands, then install the CLI with npm i -g @runcomfy/cli and sign in with runcomfy login. For CI, set RUNCOMFY_TOKEN.

Does it cost money?

It runs on RunComfy's hosted GPUs, and SKILL.md mentions being billed for GPU time and says Ctrl-C cancels the job so a stopped run is not billed. It gives no prices, so check your RunComfy account and plan before bulk runs.

Is it safe?

It runs a local CLI, sends your prompt and image URLs to RunComfy and saves files to a folder you name. SKILL.md says the sign-in token is stored in ~/.config/runcomfy/token.json with owner-only permissions, and that image URLs you pass are fetched by RunComfy's servers. Treat outside images as untrusted.

Is it made by OpenAI?

No. SKILL.md says it is not a direct OpenAI API client. It is a RunComfy skill for a model that RunComfy hosts.

Which repo is the official one?

runcomfy-com/skills is RunComfy's own GitHub organisation (the runcomfy CLI's npm package points at runcomfy-com/runcomfy-cli), and it holds a gpt-image-2 folder with the same SKILL.md. The prime-skills/runcomfy-agent-skills copy is the one behind the large skills.sh install count, and agentspace-so/runcomfy-skills is the repo SKILL.md itself names. The three installs give the same skill, so the first install command uses RunComfy's repo. The runcomfy-com/skills README says some older skill documents keep historical install links.

Does it work in the Claude app?

This page offers no Claude app upload route, because the skill needs the local runcomfy CLI. The README calls the repo "RunComfy AI media skills for Claude agents". Use it in Claude Code or another agent that can run shell commands.

What are its limits?

SKILL.md lists three fixed sizes, long embedded text that degrades output, multi-image edits that give guidance rather than strict control, and portrait photorealism that it says Nano Banana Pro does better.

Information

updated Nov 2, 2026

Repository: prime-skills/runcomfy-agent-skills
Folder: gpt-image-2

prime-skills/runcomfy-agent-skills skills.sh

Sources

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

Show all 6

skills.sh listing · skills.sh
gpt-image-2 SKILL.md · github.com
runcomfy-agent-skills README · github.com
gpt-image-2 on skills.sh (runcomfy-com/skills) · skills.sh
runcomfy-com/skills README · github.com
runcomfy-agent-skills LICENSE · github.com