Sign In

Raehoshi Moss Image

67

Updated: Nov 15, 2026

base modelanimeraehoshigame

Recently loaded

Download

2 variants available

You need these files to run this model. We'll show the best match for your preferences.

Text Encoder

qwen3.5_9b_int8_convrot.safetensors • 9.08 GB

Verified:

VAE

flux2-vae.safetensors • 320.64 MB

Verified:

Downloads your preferred variants

Type
Checkpoint Trained
Stats

233

67

35

Reviews
Published

Nov 15, 2026

Base Model
Training
Epochs: 9
Hash
AutoV265F5EA0EFB
00097-3900588371.png

Raehoshi Moss Image

Raehoshi Moss Image is an enhanced iteration built on the Moss-Image Base architecture. This release refines the visual style, adds a wide range of new concepts, and expands character knowledge. The goal is a more polished, balanced, and visually striking output that stays faithful to the core strengths of the base model.


⚠️ These settings are mandatory, especially SLG. Skipping them will noticeably degrade quality.

Sampler: ER SDE
Schedule Type: SGM Uniform
Steps: 30–50
CFG Scale: 4.0 – 5.0
Skip Layer Guidance (SLG), Single Layers: 22, 24, 26
SLG Start Percent: 0
SLG End Percent: 0.5

Where to find SLG:

  • ComfyUI: use the SkipLayersGuidanceDITSimple node.

  • Forge Neo: SLG settings are under the Stable Diffusion tab in Settings.


Version Details

v1.0

  • Dataset: trained on ~10k images, featuring the newest characters from Wuthering Waves, Zenless Zone Zero, Honkai: Star Rail, and Genshin Impact, up to October 2026.

  • Captioning: four caption formats were used for flexible prompting: full tags, natural long, medium, and short.

  • Method: built on Moss-Image Base, merged with a LoRA extracted from Moss-Image SFT (rank 128, via SVD) at the final phase of training.


Support the Development

If you enjoy this model and want to help fund future updates, consider supporting the project by buying me a coffee on Ko-fi. Thank you!


License

Released under the Apache 2.0 license.