Introduction
MOSS Cat Tower is MOSS-Image fine-tuned model.
This model features enhancements and adjustments to the default anime style.
How to Get Started with the Model
In order to perform inference, the following files are required in addition to this model.
Text Encoders (use one of them)
VAE (use one of them)
You need to use the webui that supports MOSS-Image.
ComfyUI: v0.42.1 or later
Refer here for how to place files in ComfyUI.Forge Neo: 2.31 or later (I recommend updating to at least commit 0c388f8b26d26d475c2c7104aea4eafdf7bbb3e7)
Recommended Setting
The following settings are strongly recommended (especially those related to Skip Layer Guidance). Otherwise, the art style will be inconsistent, and the body may not be drawn correctly.
Resolution: 1-2MP
Steps: 50
CFG scale: 3
Sampler: euler
Scheduler: simple
Single Layers, Layers to Skip: 22,24,26
Start Percent: 0.35
End Percent: 0.7
Configrations related to Skip Layer Guidance:
ComfyUI:
Set the values as shown in the image below to "SkipLayerGuidanceDiTSimple" Node
Forge Neo:
1.Select "Settings" -> "Stable Diffusion"
2.Set the values as shown in the image below
3.Restart Web UI
Prompting guidelines
Refer to the Prompt Guidelines of MOSS-Image-Base User Guide.
Negative Prompt
worst quality, bad anatomy, watermark, jpeg artifactDescription
This is a merged model with multiple full fine-tuned models derived from MOSS-Image-Base.
The model was trained using natural language captions that conformed to the MOSS-Image JSON format, and Danbooru tags.
Training Bucket Size: 2.4MP (1536x1536)