gpt-image-2-skill
Generate and edit images with OpenAI's GPT Image 2, from Claude Code and Codex - 21 style presets, platform sizing, carousels and photo edits. Iterate on $0.006 drafts, pay for the one you approved. A free tool by DBHQ.
<div align="center">
<img src="assets/logo.svg" alt="gpt-image-2 skill for Claude Code, by DBHQ" width="460">
gpt-image-2
Iterate on drafts, pay for the one you approved
[](LICENSE)
[](https://code.claude.com/docs/en/plugins)
[]()
A free, open-source tool by [DBHQ](https://dbhq.uk)
</div>
---
Generate and edit images with OpenAI's GPT Image 2, through a guided flow rather than a flag reference: what are we making, in which style, for where it is going, draft first, then final.
What makes it different
The draft loop is the point. A low-quality draft costs about $0.006 against roughly $0.21 for a final, so the workflow generates a draft, shows it to you, and only spends the real money once you have said yes. The approved draft's --seed carries into the final, so upgrading the quality keeps the composition you actually picked instead of rolling a new image at ten times the price. Across a ten-slide carousel that is $0.06 to find the direction rather than $2.10.
It tells you the cost before it spends it. Every run estimates first. Below $0.50 it proceeds; at or above, it stops and asks. --estimate prices a batch without generating anything, and --dry-run prints the fully assembled prompt without making a request at all. A tool that spends your money should never surprise you about how much.
One dependency, and no vendor SDK. PyYAML, for the catalogues. The API calls go out over urllib from the standard library. That is a deliberate trade: this is a skill that holds an API key, and the less third-party code sits between the key and the wire, the less there is for you to audit before you trust it.
Presets that carry the whole prompt, not a style word. 21 of them, each pairing a short description you choose from with a full prompt fragment that does the work - editorial, blueprint, ink, risograph, wireframe, constellation, brutalist, grain for visual work; infographic, slide, diagram, poster, menu, manga where the text in the image has to be legible; plus a set of community favourites. Platform sizing for the eight places images actually go.
Install
Any agent (Claude Code, Codex, Cursor, Copilot, Windsurf, Gemini, Cline and more)
``bash`
npx skills add dbhq-uk/gpt-image-2-skill
The [skills.sh](https://skills.sh) CLI installs into whichever agent directories it finds.
Local install (Claude Code or Codex)
`bash`
git clone https://github.com/dbhq-uk/gpt-image-2-skill.git
cd gpt-image-2-skill
./install.sh # Claude Code: symlinks into ~/.claude/skills (edits are live)
./install-codex.sh # Codex: installs into ~/.codex/skills
[install.sh](install.sh) and [install-codex.sh](install-codex.sh) are the same install two ways: Claude Code substitutes ${CLAUDE_SKILL_DIR} so the whole skill directory is symlinked untouched, while Codex does not, so its SKILL.md is rewritten at install time. Both build a virtualenv inside the skill directory and install PyYAML into it.
Requirements
- Python 3.9+OPENAI_API_KEY
- An OpenAI API key in . It is read from the environment and never written to disk by this skill. To route via OpenRouter instead, set OPENROUTER_API_KEY and pass --provider openrouter
- ImageMagick (optional) - needed only for platform resizing and carousel contact sheets
Usage
Describe what you want and the skill takes it from there.
`
"generate an editorial image of a rocket"
"make me a 10-slide LinkedIn carousel about spreadsheet risk"
"turn this photo into the constellation style"
"a diagram of the OAuth flow, with readable labels"
"four variants of a mountain in ink style"
Or drive the CLI directly:
`bash
PY=~/.claude/skills/gpt-image-2/.venv/bin/python
GEN=~/.claude/skills/gpt-image-2/scripts/gpt_image_2.py
$PY $GEN --draft --preset editorial "a cat astronaut" out.png # ~$0.006`
$PY $GEN --seed 42 --quality high --preset editorial "a cat astronaut" out.png
$PY $GEN --estimate --n 10 --quality high "batch test" # price it, generate nothing
$PY $GEN --dry-run --preset diagram "OAuth flow" out.png # show the prompt, call nothing
| Quality | Per image | Ten-slide carousel |--draft
|---|---|---|
| (low) | $0.006 | $0.06 |
| medium | $0.05 | $0.50 |
| high (default) | $0.21 | $2.10 |
| high + thinking | $0.25 to $0.42 | $2.50 to $4.20 |
Prices are OpenAI's at the time of writing and are baked into the estimator, so check them against OpenAI's current pricing if the numbers matter to you. The full workflow and CLI reference is in [skills/gpt-image-2/SKILL.md](skills/gpt-image-2/SKILL.md).
What this will not do
Spend money without telling you first. Every path prices the run before making it. -y exists to skip the confirmation in a batch, and it is the only way to turn that off.
Store your API key. It is read from the environment on each run. The skill writes config, a generation log and a last-run record to ~/.config/gpt-image-2/, and none of the three has a field for a key.
Development
Want to hack on the skill or run it from source with live edits? See [docs/dev-setup.md](docs/dev-setup.md).
[CONTRIBUTING.md](CONTRIBUTING.md) covers working on it, and [AGENTS.md](AGENTS.md) is for an AI agent doing so. The skill itself is [skills/gpt-image-2/SKILL.md`](skills/gpt-image-2/SKILL.md).
Acknowledgements
Adapted from [glebis/claude-skills](https://github.com/glebis/claude-skills/tree/main/gpt-image-2) (MIT), which is where the preset catalogue and the draft-then-final shape come from.
License
[MIT](LICENSE) © 2026 DBHQ Consulting Ltd
Describe what you want and the skill takes it from there.
``
"generate an editorial image of a rocket"
"make me a 10-slide LinkedIn carousel about spreadsheet risk"
"turn this photo into the constellation style"
"a diagram of the OAuth flow, with readable labels"
"four variants of a mountain in ink style"
Or drive the CLI directly:
`bash
PY=~/.claude/skills/gpt-image-2/.venv/bin/python
GEN=~/.claude/skills/gpt-image-2/scripts/gpt_image_2.py
$PY $GEN --draft --preset editorial "a cat astronaut" out.png # ~$0.006`
$PY $GEN --seed 42 --quality high --preset editorial "a cat astronaut" out.png
$PY $GEN --estimate --n 10 --quality high "batch test" # price it, generate nothing
$PY $GEN --dry-run --preset diagram "OAuth flow" out.png # show the prompt, call nothing
| Quality | Per image | Ten-slide carousel |--draft
|---|---|---|
| (low) | $0.006 | $0.06 |
| medium | $0.05 | $0.50 |
| high (default) | $0.21 | $2.10 |
| high + thinking | $0.25 to $0.42 | $2.50 to $4.20 |
Prices are OpenAI's at the time of writing and are baked into the estimator, so check them against OpenAI's current pricing if the numbers matter to you. The full workflow and CLI reference is in [skills/gpt-image-2/SKILL.md`](skills/gpt-image-2/SKILL.md).
- Python 3.9+
- An OpenAI API key in OPENAI_API_KEY. It is read from the environment and never written to disk by this skill. To route via OpenRouter instead, set OPENROUTER_API_KEY and pass --provider openrouter
- ImageMagick (optional) - needed only for platform resizing and carousel contact sheets