Look into using LLM to construct prompts conditioned on user input

[SlapDrone]

We want to avoid users having to speak prompt-engineering language. Can we use few-shot learning with an LLM and some good prompt examples / templates to offload this burden from the user and guarantee half-decent art from any description?