Currently working on Rugged resolve Z image turbo, because I cant post images on here without the specific model I have to post test pictures on the last version of RRS. so I apologize if you download any of the sd1.5 models hoping for the same results. Maybe in a week or two I'll post V1 on civitai, currently it works very well but it is still ZIT and needs a good paragraph long prompt to get real results. Male anatomy wise its fantastic, soft core and solo stuff it works great on, basic nsfw it does good on. This will not be a hardcore porn model so if you were hope for extreme penetration shots or things of that nature im sorry but as the maker of this model that stuff doesn't really interest me. currently things im still working out are better bulges, more buff burly rugged hairy men, I would still love to get both a flaccid and erect option like noob-ai and flux can do with or without a foreskin if i cant there are a couple of wonderful loras for that already on civi. So far I have achieved basically burning out female/heterosexual content this model was distilled on meaning even if wanted a naked woman she will still have a set of testicles, still working on that whole frontal penetration like you would see on a m/f shot, so far its just alot of docking so if your into that YAY. I am doing to my best to achieve that same simple but effective prompting style of https://civarchive.com/models/681901/indigo-void-furry-fused-xl-furry-noobai which I absolutely love.
Description
FAQ
Comments (17)
I'm using Easy Diffusion and get the following error: Error: Could not load the stable-diffusion model! Reason: 'Attention' object has no attribute 'to_to_k'
any suggestions to resolve this?
I've found this discussion around what causes this issue, but don't really understand the script well enough to run my own fix on the safetensor. https://github.com/huggingface/diffusers/issues/4790
Heya, Unfortunately iv never used easy diffusion So I dont know how their program works I use Stability matrix for everything. Its a all in one installer that lets you use separate packages at a time like comfy, automatic, focus, forge ect. If I can use rss on it you might want to give it a shot.https://github.com/LykosAI/StabilityMatrix
@njorthrbjorn later yesterday evening I was able to get the colab version of the code at the previously linked page to run. This resolved the error on a v7 instance of the model I was playing with. I'll try again with v9 soon.
I'm on a lower end machine and Easy Diffusion is the only interface I've had luck running at a reasonable speed (so far)
@rcbrkmyr128 i'll download easy and try that out myself and hopefully it will be resolved for future releases if others have having the same issue as well. Can you tell me where you put that source of code?
Hey, just an update I changed the code you mentioned and tried to generate a image using v10beta vs v9 released on easy diffusion and v9 gave that error while v10beta did not so you shouldent have any issues following the next release when using easydiff
If you go to download V10 be sure to download the fp8 version, its actually fp16 but civi won't let me upload 2 of the same fp files. Its a previous v10 test merge before I nuked automatic but should work well enough.
Tried both v9 and v8 files but get the following"Error: Could not load the stable-diffusion model! Reason: 'Attention' object has no attribute 'to_to_k" Four downloads and none of them will run in Easy Diffusion.
heya, another person brought that to my attention, it is apparently caused by bad coding when using automatics merging system. I have fixed the issue and on the next release easydiff will be able to use rss. Im sorry about this but just give me a day or 2 to release v10
Thank you, I actually saw your comments to the other person after making the post. I look forward to trying v10 when it's ready. Thanks for all your hard work keeping SD1 alive.
i'm just putting this here for memory or if anyone else is having this issue, but all i did was delete all the vaes in my vae folder and change the select vae from automatic to none in the settings.
If you go to download V10 be sure to download the fp8 version, its actually fp16 but civi won't let me upload 2 of the same fp files. Its a previous v10 test merge before I nuked automatic but should work well enough.
Thanks for all your hard work on this. It paid off well!
@2012shawn glad your enjoying it!
@njorthrbjorn S̶c̶r̶a̶t̶c̶h̶p̶r̶o̶o̶f̶'s stuff can now be found here. https://civitai.com/user/ctrl_alt_diffuse and https://huggingface.co/Scratchproof/Virile_Series/tree/main
@njorthrbjorn
UndefinedUsername Found the solution to the “attribute 'to_to_k” error. He states the following…
I'm not very knowledgeable in model contents. Best I could do is clean the model up with toolkit extension:
https://github.com/arenasys/stable-diffusion-webui-model-toolkit
It removed some unnecessary 'keys', whatever that means.
Just make sure that creator tests the 'clean' model in comparison to the old one. I found it can change output of some older checkpoints.
@2012shawn Thank for linking that, I didnt know that was a thing. So luckily i wasn't happy with with V10 and continued making a v11, I tried the test model on that extension and both the fp32/fp16 work in easy diff.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.