Now available for NoobAI V-Pred!
This DORA is designed to recreate the style of the artist Nayoshi. No trigger words needed, but read the usage notes and description of the score tags for details. The updated DORA version provides better hands and a much more accurate rendering of the artists style and can be used exactly as a regular LORA in most interfaces.
All sample images are straight out of ReForge UI with a Hires Fix pass. I recommend Latent Bicubic 1.5x at 0.58-0.65 denoise. No inpainting or Adetailer was used.
NSFW NOTE: This dataset contains many NSFW images. If you do not want NSFW make sure you actively prompt away from it.
Please support the original artist wherever possible so they can continue to produce their art. Without them, LORAs such as these would not be possible.
This model is also backed up on Huggingface and is available on CivitAI: https://civitai.com/models/513487/nayoshi-r-744-style
USAGE NOTES:
For both versions recommended to use full 1.0 strength.
"sketch" is a special tag that triggers the artist's older Pixiv style. You may want to include this tag in the negative for better shading and smoother outlines. You can see samples of this sub-style in the sample gallery. It can be applied to any score rating, but has best results at score_7 or score_6 (aka good quality and average quality). If going for this sub-style, you may need a higher denoise on HiRes Fix for best results.
PONY:
This version was trained on AutismMix and is recommended to use it at full 1.0 strength. If you are using a different checkpoint, you may need use a higher strength to maintain the style. Be sure to include the "source_anime" tag for best results. Use of the Expressive H LORA will result in artifacting, so please avoid using it.
Recommended positive prompt:
score_9, score_8_up, score_7_up, score_6_up, source_anime,
Use of negative embeddings will take away from the style and possibly introduce artifacts. Recommended negative prompt (adjust "hands" and "fingers" as needed):
source_pony, (hands:1.1)
NoobAI:
The NoobAI Epsilon version was trained on NAI Epsilon-pred 1.1-Version and the V-Pred version was trained on NAI V-Pred version 1.0, please use these checkpoints for best results. When using the NoobAI version, you should use the trigger word "nayoshi-style"
Recommended postive prompt:
masterpiece, best quality, nayoshi-style,
Recommended negative prompt: (you can add "hands" and "fingers" as needed)
worst quality, bad quality, screenshot, sketch, jpeg artifacts
DATASET NOTES:
The dataset for the latest version containts 515+ images. All images were hand-tagged and dewatermarked.
Score_9/"mastperiece" images are the most detailed and newest works.
Score_8/"best quality" images are slightly older or slightly less detailed works.
Score_6/"good quality" and Score_7/"average quality" images are low-detailed works and Pixiv sketches and are mostly tagged with "sketch" which produces a unique sub-style. The "sketch" tag can be applied to any score rating.