[OPENAI] Support image edits with gpt-image-1

[sbounmy]

Still a draft but I use it in production in my app https://github.com/sbounmy/hongbao_bitcoin

Usage

avatar = User.last.avatar
image = RubyLLM.edit(
  "Transform into ghibli style",
  model: "gpt-image-1",
  with: { image: [ ActiveStorage::Blob.service.path_for(avatar.key) ] } # accepts a path or remote url
  options: { size: '1024x1024', quality: 'medium' }
)

image.to_blob # image to store 
image.usage # {'input_tokens' => 362, 'input_tokens_details' => { 'image_tokens' => 323, 'text_tokens' => 39 }, 'output_tokens' => 4160, 'total_tokens' => 4522 })
image.total_cost) # 0.17002
image.input_cost) # 0.00362
image.output_cost # 0.1664

Todo :

• fix the multipart connection code vs json
• able to customize image output https://platform.openai.com/docs/guides/image-generation?image-generation-model=gpt-image-1#customize-image-output
• remove some deprecated code (errors)
• see if models.json is correct (didnt import gpt-image-1 when I ran the rake task so I added manually)

[sbounmy]

issue #138

[sbounmy]

would be great to have your feedback @crmne

I am a bit struggling with capabilities for gpt-image-1 / models.json generation

[sbounmy]

@crmne would be great to not ghost PRs. at least pin out what could be done better if you don't agree

[crmne]

Thanks for the work on this. Image editing is definitely in scope, but I'd prefer extending the existing paint method rather than adding a separate edit method:

# Generate from scratch (current behavior)
RubyLLM.paint("a sunset over mountains")

# Edit existing image (new behavior)  
RubyLLM.paint("make it more vibrant", with: "path/to/image.png")

This keeps the API consistent with how chat.ask handles attachments.

On "ghosting": I respond when I can. This is unpaid work I do between running my business and other priorities. Characterizing my delayed responses as "ghosting" is inappropriate and creates a toxic environment.

I'll review this properly when I have time.

[sbounmy]

@crmne thats the feeling I had, my bad if you were hurt by "ghosting".

As you know we also contribute (for free), when a PR doesn't get merged and we keep having conflicts as we try to catch up main branch which is frustrating.

Regarding the PR I wanted to use #paint but the edit API call involved different changes (multipart/form-data etc) that I thought might be cleaner to have a separate one. #138

Thanks for maintaining this gem

[crmne]

I appreciate the apology. The merge conflict frustration is understandable.

I still prefer extending paint rather than adding edit. The technical complexity (multipart vs JSON) should be hidden from users - that's an implementation detail. Having both methods for what's essentially the same operation (image generation) is confusing.

The API should be:

RubyLLM.paint("prompt") # generate
RubyLLM.paint("prompt", with: "path") # edit

This matches how chat.ask handles attachments and keeps the interface clean.