Sign In

DRKM4GE — Chiaroscuro | Krea 2

31

Recently loaded

Download

1 variant available

bf16 SafeTensor

DRKM4GE_Chiaroscuro_ Krea2_LoRA_step_2600.safetensors

BF16, good balance • 111.94 MB

Verified:

Type
LoRA
Stats

141

31

57

Reviews
Published

Nov 1, 2026

Base Model
Training
Steps: 2,600Epochs: 2
Usage Tips
Strength: 1
Hash
AutoV277DADE8847
Trigger Words
DRKM4GE
image (26).png

DRKM4GE — Chiaroscuro | Krea 2

Cinematic Lighting • Rich Materials • Dimensional Detail

Chiaroscuro is an experimental Krea 2 LoRA that emerged from training on 491 carefully captioned images spanning multiple artistic styles.

Originally intended to improve my existing semi-realistic anime aesthetic, the training produced something unexpected: a model that appears to emphasize dramatic lighting, dimensional shading, intricate materials, and richer surface detail.

Rather than replacing my original DRKM4GE LoRA, I'm releasing Chiaroscuro separately because it offers a different visual character.

What you can expect

  • Dramatic light and shadow with greater perceived depth.

  • Richer metallic reflections, gemstones, glass, fabric, and decorative details.

  • More dimensional character rendering, particularly in fantasy portraits.

  • Strong visual effects across both dark, atmospheric scenes and bright, colorful compositions.

  • An aesthetic that can depart from the softer appearance of my original DRKM4GE LoRA.

Training and testing

Trained on 491 images, with checkpoints evaluated from steps 200 through 4000.

I conducted a blind comparison of all 20 checkpoints against my existing production DRKM4GE LoRA across eight prompts. Step 2600 tied for the most first-place selections and appeared among my favorites most frequently.

Recommended starting settings

  • Base model: Krea 2

  • LoRA strength: 1.0 (tested)

  • Selected checkpoint: Step 2600

Lower strengths and combinations with other LoRAs have not been comprehensively validated.

Known limitations

Some generations may exhibit overly glossy skin, pronounced shadows, or changes in composition. Results depend on the prompt, seed, and workflow.

This is an artistic enhancement LoRA, not a guaranteed anatomy correction or universal detail improvement.

The story behind the model

This LoRA began as an experiment that didn't appear to improve my existing style. After extensive comparisons and blind checkpoint testing, I realized the training had developed a different aesthetic character.

Sometimes an experiment doesn't produce what you intended—but produces something worth keeping.