Building prompts with LLMs
retained text
I’ve been into AI image and video generation for a while, and I’ve been a bit obsessed with H3 for the past two months. With the right references and prompt, it feels like I can make almost anything. Writing that prompt is the tedious part, so I usually hand it to an LLM. For most of those two months, I used closed models, especially Grok (for obvious reasons). My problem is that they rarely add much on their own (no creativity). If I don’t explicitly ask for an emotion, expression, or action, they tend to leave it out. I've had generations where a character is stiff as a tree for 15 seconds which makes me redo the generation. Recently I tried Qwen 3.8 Flash Next, and it changed how I work. I give it the specifics and ask it to fill in the gaps or choose emotions that fit the scene. Sometimes the resulting clip is better than what I had in mind (I mean Oscar worthy). I’ve also started asking it to suggest the next clip, and I find its ideas more creative than those from supposedly smarter models. What do you use to write your H3 prompts?
document fetched 2026-10-12 00:29 · canonical https://reddit.com/r/StableDiffusion/comments/1x358lr/building_prompts_with_llms