[OPENAI] Support Image edits gpt-image-1

[sbounmy]

OpenAI recently released their new image generation model via API: gpt-image-1.

It would be great to add support for it in the gem.

Notes

• gpt-image-1 uses a different endpoint: images/edits.
• Currently, RubyLLM#paint uses images/generations endpoint (code).
• It's a bit odd that OpenAI introduced a separate endpoint instead of simply allowing an image input on generations, but that's the current setup.

Proposal:

I suggest introducing a new method, maybe RubyLLM#edit, to support the images/edits endpoint separately without changing the existing paint method.

Or update RubyLLM#paint to support :

RubyLLM.paint 'transform me this image into ghibli style", with: { image: some_image }

WDYT?

Lastly, I am wondering if this would be possible (instead of just #paint) :

chat = RubyLLM.chat
chat.draw "transform me into ghibli style", with: { image: "me.png" }

# Analyze images
chat.draw "can you turn the landscape into blue pastel colors?"

[gronnbeck]

According to OpenAIs documentation. The images/generations endpoint with model: gpt-image-1 will use the gpt-image-1 model instead of dall-e. So it should work out of the box without changing the code too much. Famous last words.

[sbounmy]

@gronnbeck my bad I forgot to mention that the goal is to support prompt with an uploaded image https://platform.openai.com/docs/api-reference/images/createEdit#images-createedit-image which is not supported byimage/generations

[yorsant]

It would be nice to add the new model: gpt-image-1 as an available model for images/generations.

Supporting images/edits sounds like a totally new feature.

(adding myself here to track of this issue)

[sbounmy]

@yorsant still a draft but I use it in prod #152