← sources

I made tiny sliders for MiniMax H3 that control motion, emotion, and more without training a LoRA [ComfyUI]

source: reddit (reddit:StableDiffusion)published 2026-10-30 23:52ingested 2026-10-31 02:38

retained text

Made this for the Open Source Higgsfield challenge.

ComfyUI-Dials lets you turn pairs of prompts into reusable \~14 KB sliders. No training required.

You can control motion, mood, acting style, and more. Stack them with each other or with LoRAs. They can even influence generated audio.

The video shows some of the experiments and results.

The UX is still being polished, and I'm working on better calibration for exporting dials as LoRAs. Hopefully more improvements coming soon!

**Demo:** [https://matlowai.github.io/ComfyUI-Dials/](https://matlowai.github.io/ComfyUI-Dials/)

**GitHub:** [https://github.com/matlowai/ComfyUI-Dials](https://github.com/matlowai/ComfyUI-Dials)

**Dials:** [https://huggingface.co/MATLOWAI/H3-Dials](https://huggingface.co/MATLOWAI/H3-Dials)

Curious what odd/fun things people will do. I'll adapt it for more models like MOSS/Krea here too and see how it goes.

document fetched 2026-10-31 00:31 · canonical https://reddit.com/r/StableDiffusion/comments/1xjvasi/i_made_tiny_sliders_for_minimax_h3_that_control