Well Civitai is cracking down on NSFW content. They might be slowly making it harder to see/search for this

I had the models page filtered by WAN and it appeared for me

worsk great for wan here i am using 2.1

@shura490 谢谢回答

shura490 请问下这里可以画nsfw多p的img模型有没有推荐哈

What do u mean? Pls provide the error message fom the CMD

sure Im using wan2.1-i2v-14b-480p-Q6_K.gguf and i get this, except its a much longer list, of course it may be my setup is not correct, i get this error on a couple of other loras that I use but not all

Loading Musubi Tuner format LoRA...

lora key not loaded: diffusion_model.blocks.0.cross.attn_k.lora_A.weight

lora key not loaded: diffusion_model.blocks.0.cross.attn_k.lora_B.weight

lora key not loaded: diffusion_model.blocks.0.cross.attn_k.img.lora_A.weight

lora key not loaded: diffusion_model.blocks.0.cross.attn_k.img.lora_B.weight

lora key not loaded: diffusion_model.blocks.0.cross.attn_o.lora_A.weight

lora key not loaded: diffusion_model.blocks.0.cross.attn_o.lora_B.weight

lora key not loaded: diffusion_model.blocks.0.cross.attn_q.lora_A.weight

lora key not loaded: diffusion_model.blocks.0.cross.attn_q.lora_B.weight

@mofo69 thx I am also a gguf addict. I am using usually the Q8's with multigpu on my system, 4070 12 GB. I remember the same message when I used the lactation Lora yesterday but the output was fine. I will tesst it with the Wan 720 GGUF.

@sikasolutionsworldwide709  someone else reporting same issue on another lora, different model, https://civitai.com/models/1434685/anal-missionary-wan,

@mofo69 thx yeah I wanted to get back to u after testing this but I am not sure if the Lora works that way gen msg is :Requested to load WanTEModel

loaded completely 9205.827204513549 6419.477203369141 True

Requested to load WAN21

loaded partially 8305.533077764892 8305.532470703125 0

===============================================

DisTorch Virtual VRAM Analysis

===============================================

Object Role Original(GB) Total(GB) Virt(GB)

-----------------------------------------------

cuda:0 recip 11.99GB 33.99GB +22.00GB

cpu donor 63.89GB 41.89GB -22.00GB

-----------------------------------------------

model model 16.89GB 0.00GB -22.00GB

Warning: Model size is greater than 90% of recipient VRAM. 6.10 GB of GGML Layers Offloaded Automatically to Virtual VRAM. So nothing about Loras. Wan is a good model from my test now I can say it can handle lactating excellent without lora Process is still running

@mofo69 native lora loader with clip etc is not working as suggested there and can be seen in OPs wflows. UI can check the post ans subs by honryindian, https://civitai.com/models/1399040/breast-squeeze-and-lactation-milk-spray-milking-wan-21-i2vt2v-model

@mofo69 Imhave converted the format and now testing https://github.com/kohya-ss/musubi-tuner there is a convert option u can just download the convert.py and the util folder from the repo

@sikasolutionsworldwide709  so i recently discovered if I use power lora loader by rgthree i dont get the error ,

@mofo69 U will not get the error but the lora will have no effect. No comfy lora node can handle the musubi tuner format thats why the repo provides a conversion tool.

@mofo69 I have converted the so my cmd shows. [DisTorch] Full allocation string: #cuda:0;22.0;cpu

Loading Diffusers format LoRA...

Loading Diffusers format LoRA...

Loading Diffusers format LoRA...

As for conversion, command is python convert_lora.py --input path/to/musubi_lora.safetensors --output path/to/another_format.safetensors --target other its in the repo only the convert.py and the util folder is necessary https://github.com/kohya-ss/musubi-tuner No loading diffusers message=no lora effect

The naming discipline in Lora scene is extremely poor. A point of disappointment and disgust.