Gpt Image

gpt-image: image2-workbench Skill

When to use this Skill

• The user asks for "a slide", "a figure", "a mockup", "a poster",

"a cover", "a storyboard", "a character sheet", or any other still image deliverable that suits gpt-image-2.

• The user wants bilingual (e.g. Chinese + English) typography on

the image, a domain that gpt-image-2 handles unusually well.

• The user provides one or more reference images and wants an edit

with a specific "preserve" list (keep face, swap background, etc.).

• The user wants a multi-panel artifact (storyboard, comic page,

step-by-step explainer) where panel-to-panel continuity matters.

• The user only has access to web ChatGPT and wants a maximally

detailed prompt to paste in by hand.

• The user wants a quick cost estimate or capability probe before

committing to a render.

When NOT to use this Skill

• Real-time video, animation, or any moving-image deliverable —

gpt-image-2 produces stills only.

• Transparent-background PNGs — gpt-image-2 does not support

transparent output. Recommend a different tool or post-processing.

• Pure text-only outputs (essays, captions, scripts) — answer with

text directly, do not invoke the image pipeline.

How this Skill works

The Skill is a thin wrapper around the i2w CLI from a checked-out image2-workbench repository. v0.2 is a GitHub/source-checkout release, not a PyPI/wheel release. The script scripts/run_skill.py (PEP 723 inline metadata) invokes the CLI with the right verb based on the user's request.

If i2w is not on PATH, ask the user to install it from the checkout:

pip install -e ".[dev]"

Available verbs (call via i2w)

• i2w catalog list / i2w catalog search "" --domain — find a template.
• i2w template render --vars — produce a paste-ready prompt (no API call).
• i2w render generate / i2w render edit — call the API.
• i2w batch sweep — run a quality sweep across parameters.
• i2w eval run — run an eval rubric against rendered outputs.
• i2w cost estimate — token-based cost estimate before rendering.
• i2w doctor capabilities — probe model capabilities (e.g. thinking).
• i2w preflight — validate a rendered prompt before spending.
• i2w ledger query — inspect local run history and cost/error summaries.
• i2w gallery build — rebuild the bilingual prompt gallery.
• i2w version — print the installed workbench version.

Recommended workflow

1. Identify the user's domain (business / academic / uiux / anime / poster).
2. Run i2w catalog search "" --domain to find candidate templates.
3. Run i2w template render --lang ... --vars ... --out prompt.md

to compile a prompt.

1. Optionally run i2w preflight prompt.md and i2w cost estimate

before spending.

1. If the user has an API key, run `i2w render generate --prompt-file

prompt.md ... (or i2w render edit --prompt-file prompt.md --image reference.png ...` for reference-image edits) to produce the image.

1. Otherwise, hand the rendered prompt back to the user — they can

paste it into web ChatGPT directly. This is the L3 form factor.

1. For multi-variant exploration use i2w batch sweep; for quality

gates use i2w eval run.

Constraints (must honor)

• gpt-image-2 does not support transparent backgrounds.
• gpt-image-2 does not accept input_fidelity (deprecated parameter).
• moderation defaults to auto, NOT low. Do not set low unless

the user explicitly asks for it.

• size must be multiples of 16, with max edge 3840 and aspect ratio

no wider than 3:1. Sizes above 2560×1440 are experimental and may fail or degrade.

• thinking is a probed capability, not assumed — call

i2w doctor capabilities first if uncertain.

Examples

• "Make me a SWOT slide for our Q3 review, English headers, Chinese

body." → i2w catalog search swot --domain business → i2w template render business_swot_card --lang zh-CN --vars templates/business/_vars_examples/swot_acme.yml --out swot_prompt.md → i2w render generate --prompt-file swot_prompt.md --size 1536x1024 --quality high --out out/swot.png --template-id business_swot_card.

• "Make a product analytics dashboard mockup for a design review."

→ i2w catalog search "web dashboard" --domain uiux → i2w template render uiux_web_dashboard --lang en --vars templates/uiux/_vars_examples/web_dashboard_demo.yml --out dashboard_prompt.md → i2w render generate --prompt-file dashboard_prompt.md --size 1920x1088 --quality high --out out/dashboard.png --template-id uiux_web_dashboard.

• "Edit a reference image using a rendered prompt and keep the subject

identity stable." → i2w template render uiux_design_system_card --lang en --vars templates/uiux/_vars_examples/design_system_demo.yml --out edit_prompt.md → i2w render edit --prompt-file edit_prompt.md --image reference.png --out out/reference_edit.png --template-id uiux_design_system_card.

• "I just want a paste-ready prompt for an 8-panel manga storyboard."

→ i2w catalog search "manga storyboard" --domain anime → i2w template render anime_comic_8panel --lang en --vars templates/anime/_vars_examples/comic_demo.yml --out comic_prompt.md (no API call; prompt returned for the user to paste into web ChatGPT).

Provenance

This Skill bundle and its prompts are licensed under CC BY 4.0. The underlying image2-workbench CLI is licensed under Apache-2.0. See the repository NOTICE for full attribution.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: StartripAI
• Source: StartripAI/gpt-image-2.0-workbench
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.