Downloads · 30 days
0
lilu-krea2-lora
Zaytron40k/lilu-krea2-lora
lilu-krea2-lora is a text-to-image model from Zaytron40k. Use it when you need an image from a text prompt.
Krea 2 character LoRA — trigger l1lu woman. Trained on Krea 2 RAW, intended for inference on Krea 2 Turbo.
Access
Public
Updated Sep 7, 2026
Repo size
3.9 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors2.1 GB · 100%
At a glance
- Task
- Text-to-Image
- Access
- Public
- Created
- Aug 26, 2026
- Updated
- Sep 7, 2026
- SHA
- c9ab4233
Base models
Model card
From the Hugging Face model README
lilu-krea2-lora
Krea 2 character LoRA — trigger l1lu woman. Trained on Krea 2 RAW, intended for inference on Krea 2 Turbo.
Status: ✅ training complete
Training configuration
| Field | Value |
|---|---|
| Base model (train) | krea/Krea-2-Raw (raw.safetensors, 12B DiT, undistilled) |
| Inference target | krea/Krea-2-Turbo (8-step distilled) — RAW-train / Turbo-infer |
| VAE | Qwen-Image VAE (qwen_image_vae.safetensors) |
| Text encoder | Qwen3-VL-4B (qwen3vl_4b_bf16.safetensors) |
| Trainer | Musubi Tuner (kohya-ss), networks.lora_krea2 |
| LoRA | rank/dim 24, alpha 24, all Linear layers of the DiT |
| Optimizer | adamw8bit, lr 1e-4 |
| Precision | bf16, gradient checkpointing, SDPA |
| Timestep sampling | krea2_shift (resolution-aware), weighting_scheme none |
| Trigger | l1lu woman (skin + blonde hair (twin-tails) + purple eyes + face baked into the trigger; clothing described) |
| Dataset | 63 images + paired NL captions (Seedream-generated, Western-animation style); black cami crop top + pleated mini skirt + pose/expression/framing/background described, identity in trigger. batch 1, 1024 multi-res buckets, num_repeats 2 |
| Schedule | 12 epochs = 1512 steps, save every 2 epochs, seed 42 |
| Hardware | 1x NVIDIA H100 80GB (RunPod, EU-NL-1, Musubi volume) |
Loss per checkpoint
avr_loss is a rolling average — flow-matching loss is near-flat, so use it as a sanity check and pick the best checkpoint visually.
| checkpoint | epoch | step | avr_loss |
|---|---|---|---|
-000002 | 2 | 252 | 0.0505 |
-000004 | 4 | 504 | 0.0465 |
-000006 | 6 | 756 | 0.0427 |
-000008 | 8 | 1008 | 0.0459 |
-000010 | 10 | 1260 | 0.0431 |
| final | 12 | 1512 | 0.0436 |
min avr_loss 0.0399 · final 0.0436
Inference (Krea 2 Turbo, Musubi Tuner)
Prepend the trigger l1lu woman to the prompt. Recommended: stack over the house style LoRA.
python src/musubi_tuner/krea2_generate_image.py \
"l1lu woman, <scene>" \
--dit turbo.safetensors --vae qwen_image_vae.safetensors \
--text_encoder qwen3vl_4b_bf16.safetensors \
--steps 8 --guidance_scale 1 --mu 1.15 --width 1024 --height 1280 \
--attn_mode torch --lora_weight <checkpoint>.safetensors --lora_multiplier 1.0
Dataset samples
Representative image+caption pairs from the training set:
Lilu-portrait-001-seedream.jpg
l1lu woman, close-up portrait facing the camera directly, closed lips and a calm serious stare, black cami top straps visible on her shoulders, plain white studio background.
Lilu-medium-001-seedream.jpg
l1lu woman, medium shot, standing facing forward with both hands resting on her hips, elbows out, wearing a black spaghetti-strap cami crop top and a black pleated mini skirt, lips slightly parted with a calm confident expression, plain lavender purple background.
Auto-generated by training watchdog.
- Task
- Text-to-Image
- Created
- Aug 26, 2026
- Updated
- Sep 7, 2026