Nlxl v0.8
Nlxl v0.8 is a checkpoint that produces more stable and higher quality results through enhanced Model Merge compared to v0.7.
---
Nlxl
Nlxl is a checkpoint created through model merging, combining the natural image generation capabilities of Illustrious, the extensive dataset of Noob, and the unique strengths of different models depending on the version.
Strength
Excellent artists reflection
In a personal test, I tested the reflection of the drawing styles of a total of 3999 artists from pages 1 to 200 of the Danturu artists tag. While merging is not magic, surprisingly, the artists reflection itself boasted the most similar expressiveness to existing artists among the five merging models.
Improved Natural Language Prompt Recognition
A disadvantage of models trained with booru tags is their relatively low natural language recognition rate. Because of this, there are cases where even basic natural language prompt cannot be implemented, but recognition has been improved through merging.
More Hand-Drawn Style
By utilizing Illustrious' Clip in the merging process, the basic drawing style present in WAI has been removed, allowing for the creation of more natural-looking drawings.
Recommended settings
Positive: It is best to use it with natural language based on the Booru tags, and it is strongly recommended to use the artist tag.
Negative: worst quality, low quality, bad anatomy, bad hands, bad feet, extra digit, fewer digits, watermark as default, add low quality artists. Example) worst quality, low quality, bad anatomy, bad hands, bad feet, extra digit, fewer digits, watermark, chihuri, bb (baalbuddy), iesupa
CFG: Recommended value is 5
Sampler: Euler a
Resolution: 1024x1024 | SDXL resolutions
Steps: 28024 | SDXL resolutions
Steps: 28