Fluxscale v1 ComfyUI Workflow

Flux1 is an amazing model both on the visual and prompt adherence front, but it often times lack realistic skin detail and fabric details.

Enter noise injection. TLDR; by essentially splitting the sigmas midstream and injecting a dose of noise, we can effectively mitigate the smooth skin syndrome and bring the characters to real life. It works wonders on the base gen, but also on subsequent upscales! 🙀

On an RTX3090, each 1.5K gen takes a 1mn at 25 steps, and the 3K upscale clocks at 2 minutes. I haven't tried 6K yet, though I left the group in the workflow in case you want to give it a go. Let me know how it goes and what settings you end up using!