lodestones

/

MOSS-Image-SFT-8-steps-dmd

Like 218

i will if you need me to? i have my environment already setup for the arch

you can just use the lora as is
load it in your lora loader

https://huggingface.co/lodestones/MOSS-Image-SFT-8-steps-dmd/blob/main/lora.safetensors

Thanks, Lora is working now—8 steps.

@lodestones i wonder why the 4 steps can not be loader by a lora loader, and they seem to rely on the time_embedding.safetensors and dataspace.safetensors

how are you guys using it, can anyone share workflow with lora loaded, cause i cant load or work with it all i am getting is smudge, please someone help.

i have a gguf workflow but youll need my nodes

https://huggingface.co/realrebelai/MOSS-Image_GGUFs/tree/main

i have a gguf workflow but youll need my nodes

https://huggingface.co/realrebelai/MOSS-Image_GGUFs/tree/main

can i use your workflow without using gguf models and instead opting for your w4 series

Yes it will work with the w4a8s, I just dont know about bf16 or int8 because my focus is always gguf and w4a8

Yes it will work with the w4a8s, I just dont know about bf16 or int8 because my focus is always gguf and w4a8

can you release moss image w4 version or you have already done, because i know you're fast, also what should i use as edit and nsfw gen model from flux 2 klein 4b, flux3, qwen2.1 , moss, krea2, zimage, idk why but zimage turbo feels better than krea2 as krea doesn't do nsfw much and krea2 models are so big i cant find any smaller than 7 gb , and i am trying my best to avoid ggufs just sticking to int4 int8 w4a4 w6a8 versions thanks.

https://huggingface.co/realrebelai/MOSS-Image_GGUFs/blob/main/MOSS-Image-SFT-W4A8-ConvRot.safetensors

nodes are updated to run bf16 or int variations now. LoRA support is also active as well.