How to handle task-related parameters for Drivers?

[shhlife]

I have a general question about how to approach things like audio/image generation drivers that have settings that feel more like individual task settings other than driver settings.

For example, right now when you create a prompt driver you define everything that is required to use that particular model - and then you send all the parameters to affect the output as part of the task (PromptTask). For an ImageGenerationDriver you do the same... except you also set the resolution of the image... that feels like it should be a task-related thing, not a model setting thing.

For audio tools like elevenlabs text-to-speech they have some parameters like previous_text and next_text that help shape the request. Those are definitely more task-related, but appear like they'd be things you need to set on the driver, right?

Are we handling these types of parameters the right way? do they belong on the driver? should they be part of the task? How would you give an imagegeneration tool to an agent with a driver for a particular model - but then have the agent determine what the size of the output image should be?

Let the discussions begin! :)

[collindutter]

Generally speaking we need to do a better job at allowing users to override class field values at runtime. A simple example: how would a user override the temperature of a Prompt Driver run without creating an entirely new prompt driver?