gpt-image（gpt 图像）

gpt-image

General image generation/editing CLI for OpenAI's gpt-image-2. Designed for skill-capable agents: all API parameters are first-class flags, defaults are sane, output is a file on disk. Use it whenever a runtime or user asks for image generation, reference-image editing, inpainting, dense typography, or gallery-informed prompt drafting.

One-line usage

# As an agentic skill plugin (installed through your runtime/plugin manager):
uv run "$CLAUDE_PLUGIN_ROOT/skills/gpt-image/scripts/generate.py" -p "PROMPT" [-f OUT] [-i REF...] [-m MASK] [options]

# As a direct CLI (installed via uvx or uv tool install):
uvx --from git+https://github.com/wuyoscar/gpt_image_2_skill gpt-image -p "PROMPT" [-f OUT] [-i REF...] [-m MASK] [options]

# Or once installed globally:
gpt-image -p "PROMPT" [-f OUT] [-i REF...] [-m MASK] [options]

Reads OPENAI_API_KEY from env. Writes to OUT (or auto-named YYYY-MM-DD-HH-MM-SS-<slug>.png in ./fig/ or cwd). Prints output path(s) on stdout. Exit 0 on success, 1 on API error, 2 on bad args / missing key.

CLI flags (complete reference)

Budget / quality policy for agents

Use --quality as the budget dial. There is no separate --budget flag in this CLI.

• low — cheap draft mode. Use for broad prompt exploration, collecting many variants, gallery mining, rough composition checks, or when the user explicitly wants low cost / fast iteration.
• medium — balanced mode. Use for normal one-off exploration, style probing, or cases where readability matters but the output is not yet final.
• high — shipping / report mode. Use for Chinese text, posters, infographics, paper figures, dense labels, multi-panel layouts, banners, or any asset likely to be kept.

Rule of thumb for autonomous agents:

• If the user asks for many variants, cheap, draft, explore, or collect, start with low.
• If the user asks for polished but still exploratory, use medium.
• If the user asks for final, fancy, hero, paper figure, poster, diagram, or exact text, use high.
• If unsure, keep the CLI default high for text-heavy / delivery-facing outputs; otherwise prefer medium during exploration.

Endpoint selection (official OpenAI cookbook pattern)

Both endpoints accept gpt-image-2 as of April 2026 — verified against OpenAI's official cookbook prompting guide. The skill uses the official openai Python SDK under the hood (from openai import OpenAI; client.images.generate(...) / client.images.edit(...)) — the CLI is a thin wrapper that exposes every SDK parameter as a flag.

Content policy: gpt-image-2 enforces its own content rules on the edits endpoint. Real-person-likeness edits usually refuse (400 error with a moderation message). The skill surfaces the response body verbatim on stderr and exits 1.

Canonical examples

# 1. Vanilla generate, 1K square, auto quality
gpt-image -p "a photorealistic convenience store at 10pm"

# 2. 2K portrait poster with exact Chinese text, high quality
gpt-image \
  -p 'Design a 3:4 tea poster. Exact copy: "山川茶事" / "冷泡系列" / "中杯 16 元"' \
  --size portrait --quality high -f poster.png

# 3. 4-image grid, transparent background disabled, webp
gpt-image -p "isometric furniture, minimalist" \
  -n 4 --background opaque --format webp --compression 85

# 4. Edit / colorize existing image
gpt-image -p "colorize this manga page and translate to Chinese" \
  -i page.jpg -f colored.png

# 5. Multi-reference brand collab
gpt-image -p "77 (the cat) × KFC employee poster" \
  -i cat.png -i kfc_logo.png -f collab.png --size portrait

# 6. Masked inpaint — replace sky only
gpt-image -p "replace sky with aurora, keep foreground intact" \
  -i photo.jpg -m sky_mask.png -f aurora.png --quality high

# 7. 4K widescreen render
gpt-image -p "cinematic Shanghai skyline at dusk" \
  --size 4k --quality high -f skyline.png

Response handling

• API returns data: [{b64_json: "…"}] by default; the script decodes base64 and writes bytes.
• If the API returns url instead, the script GETs the URL and writes the downloaded bytes.
• With -n > 1, files are suffixed: out.png → out_0.png, out_1.png, …

Error surface

When an agent hits exit 1, it should surface the response body verbatim — it usually names the problem (rate limit, moderation block, invalid size).

Size picking guide

Scale references (load for prompt leverage, not just CLI calls)

The CLI is only the execution layer. The main value of this skill is the prompt Scale: agents should use the reference files to read concrete gallery patterns before drafting or generating images.

• references/gallery.md — routing index for the split 162-prompt Reference Gallery Atlas. The Reference Gallery is canonical; README is only a selected visual showcase. It maps each category to references/gallery-<category>.md; load the index first, then only the relevant category file(s) to avoid context bloat. Use this when the user asks for a style/category we already cover, wants diverse ideas, asks to extend the gallery, or needs a prompt that should inherit the repo's collected taste.
• references/craft.md — expanded 19-section prompt-writing checklist distilled from the gallery: Scale-first usage, exact text, JSON/config-style prompts, fixed-region infographics, data visualization mini-schemas, research/diagram grammar, UI specs, multi-panel consistency, reference-based unlocks, three-glances test, edit invariants, dense text, and category mini-schemas. Load this when improving or debugging a prompt.
• references/openai-cookbook.md — verbatim Markdown capture of OpenAI's official GPT Image prompting guide. Load this when the user asks about official parameter semantics, endpoint behavior, migration, or model capabilities.

Reference loading policy:

• For generation/editing requests, first read references/gallery.md only as the category index, then load the relevant references/gallery-<category>.md file(s); do not load every category and do not rely on generic one-line prompting if the atlas contains a matching pattern.
• Use craft.md to refine structure after selecting a gallery pattern.
• Use openai-cookbook.md for authoritative API/model questions or when the gallery does not cover the requested workflow.
• Preserve Curated versus Author + Source metadata when adapting examples. Add new collected prompts to the Reference Gallery first; promote only representative examples to README.

Attribution

License: CC BY 4.0 for prompt patterns and gallery documentation unless a referenced upstream source states otherwise. Prompt patterns are curated from the original repo gallery, OpenAI Cookbook, and community prompt collections credited in the README. Individual source metadata is preserved per entry in the split references/gallery-*.md category files where applicable.