Update imagegen skill and fallback CLI for GPT Image 2.5

[coygeek]

Request

Please update the imagegen skill and its fallback CLI for ChatGPT Images 2.5, or point users to the maintained replacement if updates now belong elsewhere. The published helper still defaults to gpt-image-1.5 and rejects options documented for the new models.

I understand this repository's README now marks it deprecated and directs users to openai/plugins. However, the skill remains published here, and Codex also ships an embedded copy. Clear ownership and migration guidance would help users avoid installing an older version over working compatibility changes.

Current source behavior

Inspected on September 9, 2026, at repository revision 49f948faa9258a0c61caceaf225e179651397431:

• scripts/image_gen.py sets DEFAULT_MODEL = "gpt-image-1.5".
• Its quality validation accepts only low, medium, high, and auto, so it rejects the new xhigh and max values locally.
• Its size validation accepts only 1024x1024, 1536x1024, 1024x1536, and auto, rather than the currently documented custom dimensions.
• The model-prefix check accepts strings beginning with gpt-image-, so this is not a claim that every request using a 2.5 model name fails. The defaults, validation, and guidance need to agree with current capabilities.

These findings come from static inspection of the public source. No paid image-generation request was made for this report, and successful generation or account eligibility is not claimed.

Expected update

1. Document gpt-image-2.5-sunburst and gpt-image-2.5-flare, explain their intended uses, and refresh the fallback CLI's default or explicitly justify retaining an older model.
2. Accept supported quality settings and custom dimensions with validation appropriate to the selected model, preserving compatibility with older models.
3. Update the skill's CLI/API references and examples together with the script.
4. Keep the distinction between Codex's built-in image tool and the API fallback clear. API parameters should not be advertised as inputs to the built-in tool unless its actual schema supports them.
5. Identify the maintained source and coordinate with the embedded copy in openai/codex, or transfer this issue to the appropriate repository.

Official references and related work

• Introducing ChatGPT Images 2.5.
• Current API image-generation guide: documents both 2.5 models, xhigh/max, and custom dimensions.
• openai-python #3824: merged SDK support for GPT Image 2.5 models and image options.
• Previous imagegen model-default issue #363: concerned the earlier transition to Image 2, not 2.5.
• openai/codex #43965: effective-model visibility and selection for the built-in tool; related, but a different scope from this skill/CLI update.

[AlwaysFocus]

FWIW what I decided to do for the time being is just record a skill that uses computer use to drive ChatGPT through the browser which does have access to gpt-image-2.5 so that I can still drive it from a codex thread.

[coygeek]

Current 2 workarounds to access GPT Images 2.5:

1. On web go to https://chatgpt.com/images/
   Pros: Can select reasoning level
   Cons: No ability to select Sunburst / Flare.
2. On mobile iOS, select 'Images'
   Cons: No ability to select Sunburst / Flare OR reasoning level.

Not working yet:

1. MacOS app.
   Note: I think you can prompt for it, using certain keywords, but there's no dedicated 'Images' tab (like web or mobile).