Image to Image capability within SK

[arafattehsin]

Hi folks!

I know we've got a few abstractions like below:

However, considering the launch of GPT-Image-1, I think it'd be great to have the support for Image to Image as this will unlock a lot of use-cases with SK.

I am not sure if it's already there so before raising it as a Feature Request or Idea, I thought to ask here if anyone has achieved it yet.

[sophialagerkranspandey]

Tagging @RogerBarreto

[RogerBarreto]

@arafattehsin That's a good point, we don't have an Image-to-Image specific abstraction. Although thinking a bit more deeply about this, we can already achieve a Image-to-Image generation implementing a ChatCompletion interface, where the UserMessage can have the prompt and the image, and the resulting Assistant Message could also contain the generated image, I would investigate this custom Chat Completion Service approach before introducing a dedicated interface for Image to Image.

[arafattehsin]

Thanks @RogerBarreto - any working example would be fantastic, esp. with gpt-image-1 as that's the main model we are targeting.