[Workers AI] Fix gpt-image-2 input field: image -> images

[superhighfives]

The docs for openai/gpt-image-2 show the edit input as image (singular), but the runtime rejects that with:

Valid fields: prompt, images, quality, size, background, output_format

So users following the docs hit a 400 immediately. Reported by Craig Dennis after Clanker copied the example verbatim.

This updates src/content/catalog-models/openai-gpt-image-2.json so the input schema, the Image Edit example, and the TypeScript snippet all use images as an array (up to 16 entries, base64 strings or data: URIs). Also adds a Multi-Image Edit example to show the array form.

• input schema: image: string -> images: string[] (maxItems: 16)
• Image Edit example + code snippet updated to pass images: [...]
• new Multi-Image Edit example demonstrating multi-reference composition

Result on the rendered page:

const response = await env.AI.run('openai/gpt-image-2', {
  prompt: 'Transform this cartoon smiley into a photorealistic 3D clay sculpture...',
  images: ['data:image/png;base64,...'],
}, { gateway: { id: 'default' } })

Heads up - this file is regenerated by bin/fetch-catalog-models.ts from the Unified Catalog API, so the upstream catalog also needs the same fix or the next sync will revert this. CCing catalog owners per CODEOWNERS so they can chase that down: @abhishekkankani @palashgo @thebongy @roerohan @kathayl.

[github-actions]

This pull request requires reviews from CODEOWNERS as it changes files that match the following patterns:

Pattern	Owners
/src/content/catalog-models/	@abhishekkankani, @palashgo, @thebongy, @roerohan, @kathayl, @mchenco, @zeke, @superhighfives, @bfirsh, @mattrothenberg, @ethulia, @cloudflare/content-engineering, @cloudflare/pcx-technical-writing, @cloudflare/product-owners

[Copilot]

Pull request overview

Fixes the openai/gpt-image-2 catalog entry so “image edit” requests match the Workers AI runtime contract (which expects images rather than image), preventing immediate 400s when users follow the docs.

Changes:

• Update the Image Edit example and TypeScript snippet to pass images: [...] instead of image: ....
• Add a new “Multi-Image Edit” example demonstrating multiple reference images.
• Update the input JSON schema from image: string to images: string[] with maxItems: 16 (and reformat some arrays inline).

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[github-actions]

Preview URL: https://3d5eb183.preview.developers.cloudflare.com
Preview Branch URL: https://fix-gpt-image-2-images-field.preview.developers.cloudflare.com