lisa-krea2-lora

Krea 2 character LoRA — trigger l1sa 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	l1sa woman (skin + red hair (bob) + brown eyes + face + shoulder tattoo baked into the trigger; clothing described)
Dataset	63 images + NL captions (Seedream, Western-animation style); black cami crop top + denim shorts + 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	0.0443
-000004	4	504	0.0405
-000006	6	756	0.0373
-000008	8	1008	0.0400
-000010	10	1260	0.0373
final	12	1512	0.0377

min avr_loss 0.0346 · final 0.0377

Inference (Krea 2 Turbo, Musubi Tuner)

Prepend the trigger l1sa woman to the prompt. Recommended: stack over the house style LoRA.

python src/musubi_tuner/krea2_generate_image.py \
    "l1sa 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:

Lisa-portrait-001-seedream.jpg

l1sa woman, close-up portrait, facing the camera directly, wearing a black cami top with thin straps, lips slightly parted, a sultry direct gaze with a hint of a smile, plain white background.

Lisa-medium-001-seedream.jpg

l1sa woman, medium shot, standing with both hands on her hips, wearing a black cami crop top and light-wash ripped denim shorts, small stud earrings, direct confident gaze, plain lavender background.

---

Auto-generated by training watchdog.