Work towards a user interface concept for variations

[MischaPanch]

A user should not need to know details of prompting strategies. We should evaluate how well "naive" prompts work (e.g. point needle north) and evaluate strategies for improving them.

One strategy could be specific additions to the prompt like "keep the semantics, don't change overall style, ..." or prompt refinement, i.e. asking an LM to revise the user prompt and make a more detailed one given the context of the task (the SVG) and some examples or an output format.

The goal of this story is to devise a set of concepts/options that can then be translated into a UI (which options, which types of input fields, etc.)