Testing Krea2 Style LoRA Training
Beautiful_Egg6188 · reddit · 2026-07-18
The author trained another style LoRA on Krea2 and shared fairly complete training parameters and results:
- Trained using the base/raw version of AIToolkit
- Total of 2220 steps, resolution set to 512/768
- Dataset contains only 37 images, each with simple captions describing only the subject and background, omitting style descriptions
- Works even without the trigger word KuvshinIlya, but it's best to include the trigger word in the prompt when mixing with other 3D styles or realistic characters
- Trained on an RTX 4070 Super + 64GB VRAM, taking about 3.5 hours
- krea 2 Raw Model with 73% offloading enabled, float8, and default settings for the rest; captions cached
The author also notes that the current version leans more feminine for characters, and a softer version trained for around 1100 steps will be released later. Testing is currently limited, but the results seem better on male characters.
More from Multimodal
- ostris pre-training "Primal VAE": 1-channel latents match modern 8x VAEs via patching — ostrisai · 2026-10-26
- Slopcore: AI-generated meme films turn low-fi slop aesthetics into a subculture — Yamapama · 2026-10-26
- TischEins, a large Midjourney SREF library, relaunches with new design after security outage — tisch_eins · 2026-10-26
- MOSS-Image-SFT 4-step distillation tops 16.5K downloads on Hugging Face — _akhaliq · 2026-10-26
- Claude Code can now edit videos: a 10-prompt workflow from raw footage to final cut — TawohAwa · 2026-10-26
- One-sentence 3D worlds turn cinematic with CineMap and TapNow — gizakdag · 2026-10-26