illustriousXL/NoobAI Stabilizer
IllustrativeXL and NoobAI are very creative models. So they often output strange/unwanted character poses and styles.
I trained this LoRA in the hope of making them more stable and natural through learning more high-quality character-focus images, instead of only relying on tons of negative prompts. And it did achieve my goal.
It also can improve the overall output quality. Because in theory it learned more excellent quality images.
For demonstration, all cover images have no negative prompts, no quality tags ("masterpiece",""worst quality", etc.), no artist or style tags.
Recommended strength: 0.4~0.8.
This LoRA does not have text encoder and is not overtrained. (Comparing with other "strong" style/concept LoRAs.) So it does not affect the creativity/diversity/style of the original model and has very good compatibility. Fewer prompts may give better results.
Trained on 1k hand-picked images that have:
Excellent quality. (≈ Wallpaper level)
Clear characters. (Clear character lines, no obstruction, no blur effect, etc.)
Natural poses. Natural body proportions. (No close-up, exaggerated art, chibi, jojo pose, etc.)
Complex and natural background. (for example: "bikini" usually means ocean/beach background, "school uniform" means classroom/school) (Most images). No abstract art backgrounds.