YellowThing LoRA (Illustrious) | Use Plant Milk 🌿 - Model Suite
A flexible Illustrious character LoRA trained for the “Yellow Thing” bounty using the Plant Milk 🌿 - Model Suite checkpoint
Features
- Strong character consistency
- Accurate blonde hair + yellow eyes
- Ahoge preserved
- Works with multiple outfits
- Good facial stability
- Flexible prompting support
- Includes support for Luna & Sol-Blanc cat prompts
Base Checkpoint
Plant Milk 🌿 - Model Suite
Recommended Trigger - yellowthing
Recommended Weight - 1.0
For more creative variations and stronger character influence, feel free to go above:
1.0 LoRA strength
Recommended Prompt
yellowthing, blonde hair, yellow eyes, ahoge, sailor outfit
Example Prompt
yellowthing, blonde hair, yellow eyes, ahoge, sailor outfit, soft anime lighting, detailed anime illustration
Notes
This LoRA was trained on a curated multi-outfit dataset to improve flexibility and avoid outfit overfitting while preserving character identity.
Supports:
- sailor outfit
- casual outfits
- dresses
- winter outfits
- cinematic scenes
Bonus:
- black cat Luna
- white cat Sol-Blanc
Cat Prompt Example
<lora:yellowthing:0.7>, yellowthing, sailor outfit, hugging cute white cat Sol-Blanc, cozy anime scene
Recommended Settings
- CFG: 5.5–7
- Steps: 28–35
- Sampler: DPM++ 2M Karras, Euler A
- Resolution: 1024x1024 , 832x1216, 1216x832
Negative Prompt
bad anatomy, extra fingers, blurry, deformed eyes, low quality, mutated hands
Description
V2 release focused on improving full body quality, facial consistency, and outfit flexibility.
Changes:
- Improved full body face rendering
- Better eye consistency and symmetry
- Added larger balanced dataset with more pose variety
- Improved outfit flexibility and environmental diversity
- Better cinematic lighting compatibility
- Cleaner anime lineart and facial structure learning
- Improved anatomy and body proportions
- Enhanced casual, elegant, fantasy, and slice-of-life outfit generation
Training Improvements:
- Better caption structure
- Improved full body tagging
- Better dataset balance between close-up, half body, and full body images
- Optimized training settings for sharper facial detail and cleaner eyes