🫧 GPT Image 2 — Pro Pack on RunComfy

runcomfy.com · GPT Image 2 text-to-image · GPT Image 2 edit · docs

GPT Image 2 is OpenAI's ChatGPT Images 2.0 model. This skill generates and edits images with GPT Image 2 hosted on the RunComfy Model API — runcomfy run openai/gpt-image-2/<endpoint> from your terminal, no OpenAI API key required.

What GPT Image 2 is

GPT Image 2 (also called ChatGPT Image 2 or just Image 2) is the second-generation OpenAI image model behind ChatGPT's Image feature, exposed here as a standalone API. Three properties make GPT Image 2 distinct:

• GPT Image 2 directive precision. GPT Image 2 follows multi-element prompts, layout cues, and embedded-text instructions more reliably than peers in 2026. When you tell GPT Image 2 to render a specific headline at a specific position, GPT Image 2 does it. This makes GPT Image 2 the right model when what's on the canvas matters more than how stylized it looks.
• GPT Image 2 in-image typography. GPT Image 2 is the strongest text-rendering image model in the catalog. GPT Image 2 reproduces quoted characters at high fidelity — short headlines, brand marks, multilingual signage (Latin, Cyrillic, Japanese kana, Arabic, etc.). When GPT Image 2 sees "AURA" in the prompt, GPT Image 2 renders the literal word.
• GPT Image 2 composition stability across iterations. GPT Image 2 holds layout stable when you change one descriptor and keep the rest of the prompt verbatim. This makes GPT Image 2 ideal for refining a single brand asset across rounds without redrafting from zero.

GPT Image 2 has two endpoints in this skill:

• GPT Image 2 text-to-image — generate a fresh image from a prompt.
• GPT Image 2 edit — modify a reference image (or compose from up to 10 references), with natural-language preservation language ("keep face identity unchanged", "keep the brand mark").

When GPT Image 2 is the right choice

Pick GPT Image 2 when any of these is true:

• The image needs embedded text — labels, signage, headlines, multilingual typography. GPT Image 2 is the strongest option.
• You're producing brand-safe e-commerce / ad / UI mockup imagery. GPT Image 2 directive precision shines here.
• You're doing iterative refinement on a single image — change one knob, keep composition stable. GPT Image 2 holds layout across rounds.
• You're producing brand-asset localization — same source asset, many language variants of the same headline. GPT Image 2 multilingual text rendering makes this fast.
• The user explicitly asked for GPT Image 2 / ChatGPT Image 2 / Image 2 / OpenAI Image 2 — route to this GPT Image 2 skill regardless.

Prerequisites

1. RunComfy CLI — npm i -g @runcomfy/cli
2. RunComfy account — runcomfy login opens a browser device-code flow.
3. CI / containers — set RUNCOMFY_TOKEN=<token> instead of runcomfy login.

Endpoints + input schema

GPT Image 2 has two endpoints, same model.

openai/gpt-image-2/text-to-image

The GPT Image 2 generation endpoint.

openai/gpt-image-2/edit

The GPT Image 2 edit endpoint.

size=auto on GPT Image 2 edit preserves input aspect ratio — strongly recommended unless you explicitly want GPT Image 2 to reframe.

How to invoke GPT Image 2

GPT Image 2 text-to-image:

runcomfy run openai/gpt-image-2/text-to-image \
  --input '{"prompt": "<GPT Image 2 prompt>", "size": "1024_1536"}' \
  --output-dir <absolute/path>

GPT Image 2 edit (single reference):

runcomfy run openai/gpt-image-2/edit \
  --input '{
    "prompt": "<GPT Image 2 edit instruction>",
    "images": ["https://..."]
  }' \
  --output-dir <absolute/path>

GPT Image 2 edit (multi-reference, up to 10):

runcomfy run openai/gpt-image-2/edit \
  --input '{
    "prompt": "compose subject from image 1 into the room from image 2; match the lighting of image 2",
    "images": ["https://...subject.jpg", "https://...room.jpg"]
  }' \
  --output-dir <absolute/path>

The CLI submits the GPT Image 2 request, polls every 2s, fetches the GPT Image 2 result, and downloads any *.runcomfy.net / *.runcomfy.com URL into --output-dir.

For pipe-friendly GPT Image 2 usage:

runcomfy --output json run openai/gpt-image-2/text-to-image \
  --input '{"prompt":"..."}' --no-wait | jq -r .request_id

Prompting GPT Image 2 — what works

GPT Image 2 responds to specific prompting patterns better than naive prose. Apply these for sharper GPT Image 2 output. Same patterns work on both GPT Image 2 text-to-image and GPT Image 2 edit.

Be explicit with subject + setting + mood for GPT Image 2. "A close-up of a matte ceramic water bottle on warm linen, soft window light, neutral background" — three concrete directives — beats "nice product photo of a bottle". GPT Image 2 reads concrete directives well.

Quote embedded text exactly for GPT Image 2. GPT Image 2 is the strongest text-rendering model in this class — but only when you put the literal characters in quotes. Long blocks of text degrade GPT Image 2 output. For multilingual typography, name the script: "Japanese kana", "Cyrillic", "Arabic right-to-left". GPT Image 2 will render accordingly.

Use compositional cues directly. "rule of thirds", "close-up", "aerial view", "centered subject", "shallow depth of field" — GPT Image 2 understands these as real directives.

Iterate one attribute at a time on GPT Image 2. When refining, change one thing per iteration (lighting OR background OR pose OR text) and keep the rest of the GPT Image 2 prompt verbatim. GPT Image 2 holds composition stable across iterations when only one knob moves.

Don't conflict GPT Image 2 instructions. "no text" + "the word 'AQUA+' on the label" is incoherent — GPT Image 2 will pick one and you don't control which.

Don't pile up styles in a GPT Image 2 prompt. "ukiyo-e + watercolor + 8K + cinematic + minimalist" cancels out in GPT Image 2 output. Pick one or two style anchors max.

For GPT Image 2 edit specifically:

• State preservation goals to GPT Image 2. "keep the person's pose and face identity unchanged", "keep the brand mark and typography on the package", "keep the overall framing". GPT Image 2 needs to know what NOT to change.
• Use directional language for GPT Image 2 spatial edits. "Move the headline from top-right to bottom-center", not "reposition the headline".
• Multi-ref GPT Image 2: number the images in the prompt — "subject from image 1, lighting and background from image 2" — and GPT Image 2 will route the cues correctly.

Where GPT Image 2 shines

Sample GPT Image 2 prompts (verified to produce strong results)

GPT Image 2 text-to-image — product hero:

A minimal hero product still life: a matte ceramic water bottle on warm linen,
soft window light, the word "AQUA+" in clean sans-serif on the label,
subtle rim highlights, e-commerce ready, 8K detail, neutral background

GPT Image 2 text-to-image — multilingual signage:

A small Tokyo café storefront at dusk, warm interior glow,
the sign reads "コーヒー" in bold Japanese kana on a wooden plaque,
shallow depth of field, rule of thirds, cinematic

GPT Image 2 edit — background swap with preservation:

Turn the background into a bright minimal white-to-soft-gray studio sweep
with gentle floor shadow; add a large headline in-image that reads
"OPEN STUDIO" in a bold clean sans-serif, high contrast, centered;
keep the main person or product, pose, and face identity unchanged

GPT Image 2 FAQ

What sizes does GPT Image 2 support? Only three: 1024_1024 (1:1), 1024_1536 (2:3 portrait), 1536_1024 (3:2 landscape). Extreme aspect ratios are auto-resized to the nearest supported GPT Image 2 size.

How many reference images can GPT Image 2 edit take? Up to 10. The first image is the primary reference for GPT Image 2; the rest provide auxiliary cues (lighting, background, style).

Can GPT Image 2 render text in any language? GPT Image 2 handles Latin, Cyrillic, Japanese kana, Korean Hangul, Arabic, Chinese, and most major scripts. Name the script in the GPT Image 2 prompt for best fidelity.

Is GPT Image 2 the same as DALL-E? No. GPT Image 2 is the successor model behind ChatGPT's Image feature; DALL-E was the predecessor line. GPT Image 2 has stronger directive precision and stronger in-image typography than the DALL-E generation.

How does GPT Image 2 compare to GPT Image (v1)? GPT Image 2 is the second-generation model. GPT Image 2 has higher fidelity, stronger text rendering, and better composition stability than GPT Image v1.

Does GPT Image 2 generate at very high resolution? GPT Image 2 outputs at the three documented sizes only. For higher-resolution upscaling, post-process the GPT Image 2 output through a separate upscaler.

How do I reproduce a GPT Image 2 output? GPT Image 2 doesn't expose a seed parameter on this endpoint. For reproducible variants, lock the prompt and re-run.

Limitations of GPT Image 2

• GPT Image 2 has only 3 fixed sizes on text-to-image (and the same 3 + auto on GPT Image 2 edit).
• GPT Image 2 prompt length caps at a few thousand tokens. Long blocks of embedded text degrade GPT Image 2 output.
• GPT Image 2 edit multi-image support is "guidance from up to 10 refs", not ControlNet-style stacks.
• Photorealism on portraits is not GPT Image 2's strongest suit — for hyperrealistic portraits, route to a different model.

Exit codes

The runcomfy CLI uses sysexits-style codes:

Full reference: docs.runcomfy.com/cli/troubleshooting.

How it works

1. The skill invokes runcomfy run openai/gpt-image-2/<endpoint> with a JSON body matching the GPT Image 2 schema.
2. The CLI POSTs to https://model-api.runcomfy.net/v1/models/openai/gpt-image-2/<endpoint> with the user's bearer token.
3. The Model API returns a request_id; the CLI polls GET .../requests/<id>/status every 2 seconds.
4. On terminal status, the CLI fetches the GPT Image 2 result and downloads any .runcomfy.net / .runcomfy.com URL into --output-dir.
5. Ctrl-C while polling cancels the GPT Image 2 request via POST .../requests/<id>/cancel.

What this skill is not

Not a direct OpenAI API client. Not a capability grant — depends on a working RunComfy account.

Security & Privacy

• Token storage: runcomfy login writes the API token to ~/.config/runcomfy/token.json with mode 0600.
• Input boundary: the GPT Image 2 prompt is passed as JSON via --input. No shell injection.
• Third-party content: image URLs are fetched by the RunComfy server. Treat external URLs as untrusted.
• Outbound endpoints: only model-api.runcomfy.net and *.runcomfy.net / *.runcomfy.com.
• Generated-file size cap: 2 GiB.