Raehoshi Illust XL
an enhanced iteration built upon the Illustrious XL model. It aims to elevate the visual style by addressing some of the limitations in the original, such as oversaturation and artifact noise. While these issues are not entirely eliminated, noticeable improvements have been made. The goal is to deliver a more polished, balanced output while staying true to the strengths of the base model.
Special Tags
Quality Tags:
- masterpiece
- best quality
- good quality
- average quality
- bad quality
- worst quality
Rating Tags:
- safe
- sensitive
- nsfw
- nsfw, explicit
Note: using rating tags in your prompts can significantly improve overall generation quality
Recommended setting
Positive prompt :
masterpiece, best quality, sensitive, absurdres
Negative prompt :
lowres, bad quality, worst quality, bad ???????, bad hands, sketch, jpeg artifacts, ugly, poorly drawn, censor, signature, watermark
Steps : 20+
CFG : 4-6
Sampler : euler a
Standard resolution :
1216 x 832, 832 x 1216, 1152 x 896, 896 x 1152, 1344 x 768, 768 x 1344, 1024 x 1024
High resolution :
1024 x 1536, 832 x 1536, 1536 x 1024, 1536 x 832
Hires.fix Setting:
- Upscaler : 4x NMKD YandereNeo XL / 4x Illustration Janai
- Hires step : 10-15
- Denoise : 0.1-0.3
Training Details
The model was developed using a two-stage fine-tuning process. In Stage 1, new series and characters were introduced into the model. Stage 2 focused on fixing issues and enhancing the overall style for improved output.
Stage 1
- Dataset : v1-31k, v2-37k, v3-34k, v4-60k, v5_v5.1-18k, v6-15k, v7-39k
- Hardware : 2x A100 80gb, v3, v4, v5, v5.1-2x H100 80gb, v7-RTX PRO 6000
- Batch size : 32
- Gradient accumulation steps : 2
- Learning rate : 6e-6
- Text encoder : 3e-6
- Epoch : 15
Stage 2
- Dataset : v1-2.5k, v2 and v3-2.3k, v4-2.5k, v5-2k, v5.1-1.8k, v6-1.5k, v7-1.7k
- Hardware : 1x A100 80gb, v7-RTX PRO 6000
- Batch size : 48
- Gradient accumulation steps : 1
- Learning rate : 3e-6, v5.1-2.5e-6, v6-adaptive
- Text encoder : disable
- Epoch : 15
List of New Series/Characters Trained:
- Zenless Zone Zero character up to v2.2
- Wuthering Waves character up to v2.6
- Honkai: Star Rail character up to v3.7
- Genshin Impact character up to v6.0
- Dandadan
- Make heroine ga oo sugiru
- Kusuriya no Hotorigoto
- Hololive from justice and dev is
- Indie Vtuber Dooby, Yuuki Sakuna, Nimi Nightmare, and S***
- 100 girlfriends who really love you
- Haite kudasai takamine-san
- Alina clover
- Fate: mrs snake
- nikke: bready
For character prompts, please refer to the Danbooru site for accurate tags and references.
License
Each version uses a different base model, so the license may vary accordingly:
- v1, v2, v2.1 → based on Illustrious v0.1 → Fair AI Public License 1.0-SD
- v3 → based on NoobAI → Fair AI Public License 1.0-SD with an additional restriction: No Commercial Use
- v4,v5.1,v6 → based on Illustrious v1.1 → original SDXL License
- v7 and later → based on Illustrious v1.1 → original SDXL License with an additional if you use it in your merges or finetune, please leave a model link or mention, It's mandatory.
