フラット塗り / Flat Color Style — Illustrious

A light flat-color style LoRA: fewer shading steps, cleaner lines, calmer colors. It pulls Illustrious-based models away from their default heavy rendering toward the look of color manga pages and doujin covers — flat, but not posterized.

No trigger word needed. Just load the LoRA. Recommended weight: 0.8 alone / 0.4 (max 0.5) when stacked with a character LoRA Trained base: Illustrious XL

Usage

1girl, (character / hair / outfit / scene),
masterpiece, best quality

Negative: grin, smug, tan, dark skin, bad quality, worst quality, worst detail, sketch

Tips (tested)

• Keep the quality tags short. masterpiece, best quality is enough. Tags that push rendering — intricate details, cinematic lighting, highly detailed, depth of field — cancel most of the effect.

• Do not write a trigger word. It is not needed, and writing one can print stray letters into the image.

• Add grin, smug, tan, dark skin to the negative. Without it the LoRA leans toward a smug grin and slightly tanned skin (a dataset bias).

• With a character LoRA, stay at 0.4–0.5. At 0.6+ some seeds start to pull hair color or add stray effects.

• Plain backgrounds (white background) come out very light grey rather than pure white.

What was verified

• Same seed, with and without the LoRA, 4 SFW scenes over several seeds: hair color tags are respected (black / blonde / silver), no extra props, no color shift at 0.8.

• Complex multi-figure poses: 4 compositions × 3 seeds, alone (0.8) and stacked with a character LoRA (0.4). No melted or fused anatomy in 24 images.

• Stacked with a character LoRA at 0.4: the character's horns, two-tone hair and accessories stay intact.

Training

43 hand-picked Danbooru flat_color illustrations from 37 different artists (max 3 per artist), chosen for white/plain backgrounds, thin lines and one-step or no shading. Only the style block of the U-Net was trained (the B-LoRA style block, up_blocks.0.attentions.1), which keeps the LoRA from changing composition, props or the character.

Caveats

• Strong composition prompts still win; the effect is a nudge, not a full repaint.

• The style is "flat-ish anime" in general, not any single artist's style.