LoRA Training Troubleshooting: Fails to Retain Outfit Details
RedDevil6064 · reddit · 2026-07-16
A developer reported generation anomalies while training a LoRA for the Krea-2-Raw model using AI Toolkit and sought help.
The Issue: After training on a specific "mouse" character, the generated images lost the outfit features from the dataset, retaining only the brown shoes.
Current Config: 600 training steps, 0.0001 learning rate, adamw8bit optimizer, and trainunet enabled. The community can use this info to diagnose whether it's an overfitting or dataset captioning issue.
More from Multimodal
- Magnific teases October 28 release: 'We're making it move' — cuenca · 2026-10-27
- Midjourney --sref demo: retro Americana Batman art series — gizakdag · 2026-10-27
- AI video is a new medium, not cinema or animation — and it's perfect for depicting dreams — Kyrannio · 2026-10-27
- Dad turns his toddler's 31-year-old quote into a song via SUNO kids' choir — NathanpmYoung · 2026-10-27
- Band's local-AI music video hits 200k views: 900 generations kept 64, 3-second clips beat face drift — Hwang0500 · 2026-10-27
- Adobe Firefly video editor walkthrough: one-sentence Quick Cut turns 32s into 20s — LudovicCreator · 2026-10-26