Using a large language model as a prompt translator for the shot prompt in the AI render workflow (Free AI renderer ComfyUI workflow: setup and single-reference run). The control passes fix the geometry, so the prompt only has to carry style and detail. An LLM can write or rewrite that part.
The author pasted the realistic prompt into Claude with the instruction to change the style to anime. The resulting video was in anime style. It also added details the author had not asked for, such as the character blinking.
A short human request is expanded by Claude (any chatbot works) into a long, complete prompt for character sheets, locations, props or video.
The human review step matters because the generated prompt can drift from the approved designs: see Asset-first approach and Audit the references before rewriting the prompt. Used in AI video inside a traditional animation pipeline.