Identity LoRA Trained on 131 Images Still Drifts on Faces — Reddit Asks for Proven ComfyUI Workflow
DragonfruitNo9822 · reddit · 2026-10-06
A user trained an identity LoRA on 131 images and runs it with Krea 2 Raw in ComfyUI, but facial proportions and cheek fullness still drift between generations.
Current setup: Krea 2 Raw, Qwen3-VL CLIP, 20 steps, CFG 1.0, Euler. The goal is maximum identity consistency while freely changing pose, clothing, lighting and background.
The author asks whether to tweak LoRA strength, reference/grounding settings, training parameters, or use an identity/reference node — and wants a proven ComfyUI workflow.
Related event: Identity LoRA Trained on 131 Images Still Struggles With Face Consistency(2 posts)→
More from Multimodal
- Prompt Share: Holographic Data Projection for Sci-Fi AI Images — azed_ai · 2026-10-06
- Open-source huashu-art-motion skill animates 35 art styles via coding agents — AlchainHust · 2026-10-06
- One Prompt Runs Claude Opus and Seedance Together for Video — Itchy-Car-904 · 2026-10-06
- MiniMax H3 Local Benchmark: How Long for 17s 720p Video on RTX 5080? — off_rs · 2026-10-06
- A Krea 2 realism workflow: Chroma for composition, LoRA for character consistency — Wide_Director_8897 · 2026-10-06
- AI-generated anime stage clip wows Reddit with smoke and lightning vibes — Amazing_Skill_6080 · 2026-10-06