模型库 /
整模型 /
MoeFussion 5.0 Moss class
文件清单 线路:镜像(hf-mirror.com) · 切换
下载后放进对应的 ComfyUI 目录。models/ 指 ComfyUI 的模型根目录。
说明
MoeFussion 5.0 Moss class
作者:LoliRimuru
MoeFusion 5 — MOSS Class A MOSS-class anime checkpoint, based on MOSS-Image and inheriting its license (Apache-2.0). What it is MoeFusion 5 is aesthetically tuned toward detailed anime. It keeps the MOSS backbone's strong natural-language understanding, so plain NLP-style prompts work great and are preferred over tag/booru-style prompting. Describe what you want like you're talking to a person. Recommended settings
| Setting | Standard | DMD variant |
|---|---|---|
| CFG | 3+ | 1 |
| Steps | ~50 | 8 |
| Sampler | Euler / ER SDE | Euler |
| Scheduler | Simple / SGM Uniform | SimpleTurbo LoRAs also work well on the standard variant if you want speed without the DMD build. |
Styles The model ships with A and B style variants — check the example images to see which fits your taste. DMD vs non-DMD
- DMD variant: 8 steps, Euler, Simple scheduler, CFG 1. Fast, and can be more stable at times — but also a bit more stiff/less flexible in output.
- Standard variant: the full-quality path. Slower, more expressive range. Status MOSS finetunes are still in an early phase, so treat this as a "good beta":
- Technically it's ahead of Krea and Anima (prompt adherence, coherence, text handling)
- Anima still wins on pure style and character knowledge — for now Credits
- Built on Kromoss by lodestone — all credit for the base fine-tune goes there
- MOSS-Image by the OpenMOSS team
