feat: Implement DALL-E image generation for OpenAI Bridge

[DZunke]

WDYT? Dall-E 3 and Dall-E 2 have the same endpoint for image generation but differ a bit in input and output. I tried to combine it, even i think on the long run, when OpenAI is not deciding to bring Dall-E 3 nearer to Dall-E 2 features, it would profit from splitting the image generation models in their specific classes for type safety but for now it should be fine, from my PoV.

It is not urgent from my side to have this feature, as i already utilize it in a slimmer version in my own code. So, if there is something to think about, just let discuss if there are general ideas for improvements here 😸

[chr-hertel]

Thanks @DZunke! Looks pretty good in first glance, will give it a try later - thanks :)

[OskarStark]

I would also say it looks good, but didn't test it

[chr-hertel]

I guess on detailed level Oskar already made some good remarks, taking one step back my first concerns would be more about OOP slicing and naming.

In general GenAI is about multi modal content input and output and I wonder what your perspective is about promoting the PhpLlm\LlmChain\Model\Message\Content one level to PhpLlm\LlmChain\Model\Content and also using it with the response instances - we already have an image class there.

[DZunke]

I guess on detailed level Oskar already made some good remarks, taking one step back my first concerns would be more about OOP slicing and naming.

In general GenAI is about multi modal content input and output and I wonder what your perspective is about promoting the PhpLlm\LlmChain\Model\Message\Content one level to PhpLlm\LlmChain\Model\Content and also using it with the response instances - we already have an image class there.

From my PoV the ImagesResponse is very special to Dall-E model because it contains the revised_prompt output made by Dall-E 3 as the model rewrites the given user prompt and it is a worth information what prompt was really taken into account for creating the image. WDYT?

[chr-hertel]

is very special to Dall-E model because it contains the revised_prompt output made by Dall-E 3 as the model rewrites the given user prompt

fair, and we can generalize later - but on the first research this is pretty unique to DallE, true

[chr-hertel]

Thanks @DZunke - nice addition :)

[OskarStark]

nice addition :)

Yes, Thanks!!!