gpt-image-cookbook

Multi-provider AI image generation cookbook — prompt gallery, agentic skill, and CLI for OpenAI gpt-image-2, Google Imagen, Flux, and more.

---

gpt-image-cookbook is an open-source multi-provider AI image generation toolkit that bundles a curated prompt gallery, an agentic skill (SKILL.md runbook for Claude Code, Codex, OpenClaw, and Hermes), and a Python CLI (gic) into one repository. It supports OpenAI gpt-image-2, Google Imagen, and Flux (fal.ai / Replicate) under a single interface for text-to-image generation, reference-image editing, inpainting, and multi-reference workflows.

If you build with AI image models, this gives you copy-paste prompts that work, a CLI that handles auth and edits, and an agent runbook that wires it all together.

---

What this is

Three things bundled together:

1. A curated prompt gallery — copy-paste prompts organized by category (posters, UI mockups, photography, diagrams, brand systems, edit/inpaint workflows) that produce reliable results across providers.
2. An agentic skill — SKILL.md runbook for Claude Code, Codex, OpenClaw, Hermes, and other skill-capable agent runtimes. Tells the agent how to search the gallery, refine the prompt, and call the CLI without writing one-off scripts.
3. A CLI (gic) — one command, multiple providers. Switch between OpenAI gpt-image-2, Google Imagen, and Flux with a single --provider flag.

---

Why a cookbook

Most prompt galleries are read-only inspiration. This one is wired into an agent that uses the gallery: search → refine → generate, with consistent semantics across providers. The CLI handles auth, encoding, sizing, edits, and inpainting so the agent never reinvents API plumbing.

---

Install

# pip
pip install gpt-image-cookbook

# or one-shot via uvx
uvx --from git+https://github.com/eugeniughelbur/gpt-image-cookbook gic --help

Set at least one provider key:

export OPENAI_API_KEY=sk-...        # for openai (default)
export GOOGLE_API_KEY=...           # for imagen
export FAL_KEY=...                  # for flux

The CLI also reads ./.env and ~/.env (without overriding env vars already set).

---

Quickstart

# Text-to-image, OpenAI default
gic -p "A minimalist conference poster, headline reads exactly 'Signal over Noise'" --quality high --size portrait

# Reference edit
gic -p "Make the sky a stormy sunset" -i ref.png

# Inpaint
gic -p "Replace the masked area with a coffee cup" -i ref.png -m mask.png

# Switch provider
gic -p "Photoreal product shot of a ceramic mug on oak" --provider imagen --quality high

Outputs land in ./generated/<timestamp>-<slug>.png unless you pass -f.

---

Use as a Claude Code plugin

# In Claude Code
/plugin install eugeniughelbur/gpt-image-cookbook

The agent loads skills/gpt-image-cookbook/SKILL.md, follows the operating loop (classify → search gallery → refine → generate), and calls gic for you.

---

Use as an agent skill (Codex, OpenClaw, Hermes, …)

Point your runtime at skills/gpt-image-cookbook/SKILL.md. Compatible runtimes auto-resolve the gic CLI via command -v gic, uv, or uvx.

---

Repo layout

.claude-plugin/        # Claude Code plugin + marketplace metadata
skills/
  gpt-image-cookbook/
    SKILL.md           # agent runbook
    references/        # gallery routing index, per-category prompts, craft cheatsheet
    scripts/           # generate.py launcher
    agents/            # runtime metadata for OpenClaw / Hermes / etc.
src/gic/               # the CLI (Python)
docs/                  # gallery thumbnails (added as you build entries)

---

CLI reference

Flag	Values	Use
-p, --prompt	string	required prompt or edit instruction
-f, --file	path	output path; auto-named if omitted
-i, --image	repeatable path	reference image; switches to edits endpoint
-m, --mask	PNG path	alpha mask for inpaint; requires -i
--provider	openai, imagen, flux	provider router
--model	string	override the provider's default model
--size	1k, 2k, 4k, portrait, landscape, square, wide, tall, or WxH	canvas size
--quality	low, medium, high, auto	cost/quality dial
-n, --n	integer	number of images
--background	auto, opaque, transparent	background mode
--format	png, jpeg, webp	output encoding
--user	string	passed to provider for end-user attribution

Exit codes: 0 success · 1 API/refusal · 2 bad args/missing key.

---

Adding a prompt to the gallery

1. Generate something you like with gic.
2. Save the preview thumbnail under docs/<category>/.
3. Add an entry to the matching skills/gpt-image-cookbook/references/gallery-<category>.md using the template documented inside that file.
4. Open a PR.

---

Adding a new provider

The provider abstraction lives in src/gic/providers/. Each provider implements Provider.generate(req: GenerateRequest) and gets registered in providers/__init__.py. See openai_provider.py for the reference shape.

---

FAQ

What is gpt-image-cookbook?

gpt-image-cookbook is an open-source toolkit for AI image generation that bundles three things in one repository: a curated prompt gallery (copy-paste prompts that work), an agentic skill (SKILL.md runbook for Claude Code, Codex, OpenClaw, and Hermes agent runtimes), and a Python CLI (gic) wrapping OpenAI gpt-image-2, Google Imagen, and Flux under one interface.

Which AI image models does it support?

OpenAI gpt-image-2 is fully implemented for text-to-image, reference-image edits, inpainting (with PNG alpha mask), and multi-reference workflows. Google Imagen (imagen-4) and Flux (flux-pro-1.1, flux-schnell via fal.ai or Replicate) have provider stubs ready — the abstraction is in place, implementations land progressively.

How do I install it?

The fastest one-shot path is uvx --from git+https://github.com/eugeniughelbur/gpt-image-cookbook gic --help. For repeated use, clone the repo and pip install -e ".[openai]". A PyPI release (pip install gpt-image-cookbook) is planned.

How do I use it as a Claude Code plugin?

Run /plugin install eugeniughelbur/gpt-image-cookbook inside Claude Code. The plugin loads the agent skill at skills/gpt-image-cookbook/SKILL.md and resolves the gic CLI automatically. The agent then follows the operating loop documented in SKILL.md (classify → search gallery → refine → generate).

What's the difference between this and just calling the OpenAI API directly?

Three things you don't get from the raw API: a gallery of working prompts organized by category, a multi-provider abstraction so you can switch between OpenAI / Imagen / Flux without rewriting code, and an agent runbook that lets a Claude Code or Codex agent use the gallery + CLI without you writing one-off scripts.

Does it cost money to use?

Yes — calls go to OpenAI / Google / fal.ai / Replicate and bill the user's account. The CLI itself is free and open-source (MIT). The cookbook recommends drafting at --quality low first ($0.01 per image on gpt-image-2) and only moving to --quality high ($0.17 per image) when the prompt is locked.

Where are prompt API keys stored?

The CLI reads keys from process environment first, then ./.env in the project, then ~/.env in the user's home directory — never overriding values already in env. Keys are never written to disk by the tool and never printed in output.

Can I add my own prompts to the gallery?

Yes — that's the point. Each category has a gallery-<category>.md file with an entry template. Generate something via gic, save the preview to docs/<category>/, paste an entry, open a PR.

How do I add a new image provider?

Implement Provider.generate(req: GenerateRequest) in a new file under src/gic/providers/ and register it in providers/__init__.py. The OpenAI implementation in openai_provider.py is the reference shape.

---

License

MIT — see LICENSE.

---

Citation

If you use this cookbook in research or tooling, please cite via CITATION.cff or:

Ghelbur, E. (2026). gpt-image-cookbook: a multi-provider AI image generation cookbook with prompt gallery, agentic skill, and CLI [Software]. https://github.com/eugeniughelbur/gpt-image-cookbook

---

Author

Built and maintained by Eugeniu Ghelbur (@eugeniughelbur).