ComfyPilotComfyPilot
模型库 / LoRA / MOSS Shoulder Width Lora
MOSS Shoulder Width Lora

MOSS Shoulder Width Lora

3059676

LoRA 21 次下载 1 喜欢 更新于 2026-11-21

底模:MOSS Image

conceptshoulder widthmoss shoulder slidershoulder slider

下载模型 在源站打开

文件清单 线路:原始(huggingface.co) · 切换

下载后放进对应的 ComfyUI 目录。models/ 指 ComfyUI 的模型根目录。

权重 moss_shoulder_sliderV1.safetensors 35 MB 下载

说明

MOSS Shoulder Width Lora

作者:viltorx294

Hello everyone, Ever since I started generating images on MOSS I wasn't completely satisfied with the way the model handles shoulders, as far as my tests showed MOSS simply thinks that: Narrow shoulders = skinny/weak Broad Shoulders = strong/bulky I really disliked that and I got tired of waiting for someone else to create a shoulder slider so I decided to make one. ATTENTION!!! I won't lie to you all and pretend this is the perfect shoulder slider (It isn't actually a slider but it behaves like one...) I'm barely satisfied with the results if I'm being honest, however, it has been around two weeks since I started this project and it has been much, much harder than I could've ever anticipated, partly due to the aforementioned way in which MOSS handles shoulders and partly due to the fact that I'm just a dude with an old 2060 6gb that works and studies... all of that to say that you should treat this as "better than nothing" more than a full fledged perfect lora, I did (and re-did sooooo many times...) the dataset using the amazing checkpoint: Juggernaut MOSS https://civitai.com/models/3042282/juggernaut-moss. and subsequently warped using scripts amongst other things, I had no idea of how to train models or even what a dataset was before this project so I also had to look at videos, look for help with AI models like Claude, etc... This first version (Again, I don't want to lie this has been exhausting and thus I can't promise that I will make another version or that I'll continue refining this LoRa more than I have already, I might but I also don't want y'all to get your hopes up...) was trained with portrait, square to the camera, studio style pics and therefore performs better with those, it struggles with more complex poses like 3/4, sideways or backgrounds and it tends to change details about the picture, specially if they are left unprompted... As I said before it isn't technically a slider since I was having a lot of trouble with AI-toolkit's "concept slider" however, it behaves exactly as one, with the safe range being -1 to 1, for a stronger effect, (although sometimes depending on the base model, prompt and seed changes more details or less) and still relatively safe it can go from -1.5 to 1.5 and finally from my tests -2 to 2 doesn't destroy or make a mess of the image however it does change it significantly relatively to 0. Unfortunately the narrow effect is much more subtle than the broad one, although, it is also relatively more stable at -1.5 and -2 than the broad one. Please let me know your thoughts and feedback in the comments, or ideas on how to make it better, if I decide to sink even more time into it... and also if you do use it please upload your photos with the tag since I am very curious to see what you do with it! I hope it behaves well in all of your generations and that someone will appreciate it more than I do, since that's the only reason I'm uploading it instead of letting it be forgotten in my pc :)

示例图