Free Krea 2 character LoRA library grows to 70 models, with full training workflow revealed
Cold-Worldliness5392 · reddit · 2026-09-25
The author expanded their free Krea 2 character LoRA library from 18 to 70 models and published the complete training pipeline, plus a visual browser showing sample generations, trigger words, recommended strengths, chosen epochs, and identity prompts.
Key workflow details
- Environment: Fizgig on RunPod for cloud LoRA training
- Datasets: large candidate pools prioritizing variety in angle/expression/lighting/hairstyle/clothing; face detection against a reference image to reject wrong-person photos; manual dedup and quality curation
- Captioning: Qwen3-VL generates descriptive captions with consistent identity triggers
- Training config: Krea 2 Standard, rank 32 / alpha 32, 64 epochs, save every 4, batch size 1, 0.25 MP target, EMA 0.98
- Selection & iteration: manually compare epochs and strengths (e.g., Angel Reese tested at 24/51/64, chose 64); weak likenesses get retrained or released as V2
A recorded Angel Reese run completed 4,416 steps across 64 epochs in 48m 46s on RunPod.
More from Multimodal
- MiniMax H3-Image image model debuts on Tencent's Miora, free to try Nov 11–17 — aziz4ai · 2026-11-11
- MiniMax H3 hits 100 days open-source: 1,472 derivative models, 148M+ downloads — linoy_tsaban · 2026-11-11
- MiniMax H3-Image explodes an espresso machine photo into 15 labeled parts — Charlietest123 · 2026-11-11
- Indie dev built a tool that clips best moments from 4-hour streams in 30 minutes — kissaadams · 2026-11-11
- Non-Blender user makes a video in 30 minutes with Claude and Seedance — FinanceYF5 · 2026-11-11
- Whimsical Woodland: A Watercolor AI Art Prompt Template for Dreamy Forest Scenes — LudovicCreator · 2026-11-11