Krea 2 LoRA training: multi-outfit characters lose likeness as Civitai's real-person purge guts the ecosystem
ArmadstheDoom · reddit · 2026-10-04
A Reddit user is struggling with Krea 2 LoRA training: basic style/character/concept LoRAs work, but results are too rigid. After Civitai removed all real-person LoRAs, the ecosystem and knowledge for training characters with specific looks or multiple outfits was decimated.
The core problem: training one Iron Man armor works, but combining multiple outfits in a dataset degrades the likeness of both. Trigger words don't seem to train anything. The poster is debating whether it's a dataset issue — do individual concepts need 20-30 images each, closer to 100 total — or whether more images simply mean the model sees each one less and learns worse, hoping to capture outfit concepts without baking them in or muddying training.
More from Multimodal
- AI Filmmaking Tip: Extract Camera Motion From Reference Clips Instead of Prompting — eyishazyer · 2026-10-05
- Dreamina's Video-to-Prompt Reverse-Engineers the Prompt Behind Any Clip — eyishazyer · 2026-10-05
- AI video: Back to the Future DeLorean flies through 100 years of history — mark_k · 2026-10-05
- Open workflow extends static video shots without degradation using Ref2v, masks and bridge resampling — roychodraws · 2026-10-05
- AI Short Puts Two Michael Jackson Personas on One Mic for a Rap Duet — SimplyAnnisa · 2026-10-04
- Kling 4.0 Fast shows surprisingly strong acting: expressions, body language and timing — LudovicCreator · 2026-10-04