“Presets”, with customisable automatic behaviour based on context

As suggested by Federico and John:

Federico’s take was that you could change model based on file attributes: size, type, name, extension. (TODO: I need to hear specific use cases to fully understand this one)

John’s take was that if you request an image to be resized for a “website” you could provide context in advance to tell the LLM that website images should always be 1400 pixels wide and proportionately resized.

This idea would be relatively straightforward prompting-wise since we could just let people add arbitrary messages to the system prompt. But for smaller LLMs they’re likely to get confused if the prompt gets too long.

One possible idea is to only append small prompt changes under certain specific circumstances:

• word tags: literally if you use the word “website”, it adds “website images should be proportionally resized to 1400px wide” to prompt. Still a bit long for a small LLM? And could confuse it if you’re trying to do something completely unrelated with the word “website”

• Similarly for other conditions: specific directory, specific file extension, etc

Any other ideas?