ai_task.generate_image: allow specifying width and height of image to generate

[candrews]

Describe the feature

I'd like ai_task.generate_image to be able optionally specify the size of the image to generate.

I can think of 2 designs for this new input data:

• 2 new optional inputs, width and height.
• Instead of width and height as a separate inputs, a size input taking values in the form of widthxheight (ex 880x524).

When specified, the image generated will be of those dimensions.

Use cases

Oftentimes, images of specific dimensions need to be generated. For example, a screen may be of 880x524 pixels and an image filling that screen is desired.

Currently, ai_task.generate_image just generates an image in unknown dimensions (some default specified in the backend configuration, unknown to the user).

Anything else?

The OpenAI specification supports a "size" input, so all OpenAI compatible implementation can implement this design. See https://developers.openai.com/api/reference/resources/images/methods/generate#(resource)%20images%20%3E%20(method)%20generate%20%3E%20(params)%200.non_streaming%20%3E%20(param)%20size%20%3E%20(schema)

[candrews]

FYI @skye-harris as this would be great to have implemented in https://github.com/skye-harris/hass_local_openai_llm/