Downloads · 30 days
0
krea2_realism_lora_v1
bieluu/krea2_realism_lora_v1
krea2_realism_lora_v1 is a machine learning model from bieluu. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as cc0-1.0.
Style LoRA for Krea 2 (trained on RAW, applied on Turbo) that pushes seamless PBR base-color texture generations toward real photographed materials.
Access
Public
Updated Sep 6, 2026
Repo size
449 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors449 MB · 100%
At a glance
- License
- cc0-1.0
- Access
- Public
- Created
- Sep 6, 2026
- Updated
- Sep 6, 2026
- SHA
- 6362e266
Base models
Model card
From the Hugging Face model README
Krea 2 realism LoRA v1 (pbrize) — no trigger word
Style LoRA for Krea 2 (trained on RAW, applied on Turbo) that pushes seamless PBR base-color texture generations toward real photographed materials.
- Data: 849 CC0 photo-derived (photogrammetry / scan / approximation) basecolor tiles from MatSynth, 1024² toroidal crops + full-sheet downscales; captions written by gpt-5.6-luna in the pbrize enhancer register.
- Trainer: fal
krea-2-trainer, 1500 steps @1024, LR 5e-4, rank 32 / alpha 32, no trigger phrase. - Keys renamed from fal's PEFT/native export to the diffusers names ComfyUI's Krea 2 loader maps
(
transformer.transformer_blocks.N.attn.to_q.lora_A.weight, …).
| file | contents |
|---|---|
krea2_realism_v1.safetensors | all trained modules (transformer blocks, text pathway, img/time/final projections) |
krea2_realism_v1_blocks_only.safetensors | transformer blocks only — text-pathway (txt_in / text_fusion) and projection deltas stripped; use this on Turbo |
Training pipeline: model-training repo, scripts/krea2_lora/.
- License
- cc0-1.0
- Created
- Sep 6, 2026
- Updated
- Sep 6, 2026