Character LoRA Fails in Krea2 Raw Mode, Developer Seeks Workflow Help

maxiedaniels · reddit · 2026-08-10

A developer trained a character LoRA for Krea2 using a Hugging Face tool. While the LoRA works perfectly in Turbo mode, switching to Raw mode and adding the Turbo LoRA (weight 0.6, 12-16 steps) causes character consistency to completely disappear. The poster is asking the community for advice on fixing this workflow issue.

Original post →

More from Multimodal

Multimodal channel →