<3 Please join my PATREON page, where you get additional content (for SDXL) and personal help, IF I can help you^^:
https://www.patreon.com/polyhedron_ai
*Publishing Notes:
Version 1.0 is a merge of four different custom LoRAs: one for female skin, one for female hands, one for male skin and - surprise! :D - one for male hands. (This version is 288 mb and in PickleTensor.)
Version 1.1 is a completely new-trained LoRA containing ALL above. And so it differs from the first version, which can be as good as bad. So: simply try BOTH, to get the output YOU desire. :) :-* // (This version is 144 mb and in SafeTensor.)
------------------------------------------------------------------------------------------------------
This LoRA has been trained to support realistic skin with blemishes as well as help create realistic hands for woman and for man. The model already works with simple resolutions up to 768px. It is actually not necessary to use an upscaler or sharpener ... Have fun with it. <3
Tipp: Get another LoRa for perfect eyes. (not made by me yet) Like this one:
https://huggingface.co/casque/detailed_eye-10/tree/main
Trivia: The example pics are re-made from Olivio Sarikas used on this tutorial:
Frank
Description
FAQ
Comments (75)
why is it in pt format?
It's all good. Made with Kohya. Did not select *.safetensors because it's home made. I can trust myself. :D Can you?
dude .. really !! ... nah no matter ..
just asking cuz I kinda misplaced some files like Loras and Lycoris once and there were some issues ..
be careful with not safetensor formats. they can basically include every python command imaginable. OP had a good idea with the model but the format he uploaded will decrease the likeliness of people downloading it immensely. Its just not worth it risking all your data getting decrypted or sold on the darkweb for a lora that can already be achieved with other .safetensor files.
@midnight_phantom the only problems are that the console will tell you that a specific embedding not works referencing the name of the file put into the wrong folder, fixed immediately by putting that file into the right folder and restarting.
But lora in .pt format is REALLY uncommon, ive personally havent seen it yet and i'd always recommend staying on the safe side and let other people test the waters for you. especially since OP is a new creator with only one model up i'd be careful downloading such files.
the pickle safety checker civitai uses can fish out some of the possible malware put onto this site but is still no guarantee a file isnt malicious. safetensors try to solve most of the security concerns by being basically just data not able to execute python code. from what i heard there is the possibility of exploiting safetensors too but that would afaik require complex zero-days.
Not saying Polyhedron_AI or this lora specific are malicious actors, just raising awareness that those formats hinder all of us and make people uploading their first model in weird formats seem pretty sus :D
as a grown up, ive gotta ask one question:
.safetensor pls?:3
@JarJar737 Aaand: It's still absolutely safe. :D
@Polyhedron_AI saying it's safe is not the same as providing people with an absolutely safe file and .safetensors ARE indeed safe.
@DarkAgent Yes, I know. I simply did not pay attention to that problem. I MADE this LoRA. I used this model for weeks on my computer ...
@Polyhedron_AI please, hear the people. Safetensor is a standard format, no pt.
@Polyhedron_AI It doesn't matter how much you claim it to be safe. Would you trust me if I sent you a suspicious file and saying "trust me bro, it's safe"? If you want to share your creations, upload it in the appropriate file format.
@bomber055 Same answer here: If you find out how to convert this existing *.pt file into safe tensors without re-training I would do it: It's a merge of FOUR different custom LoRAs. I could not simply do it completely new, I'd would change. :)
@Polyhedron_AI So, now you have an experience which file format to use next time. )
What about the example of the hands? I think peoples care more about hands than skin
Try out for yourself. :D Or check my FB-Page. :)
i dont want to use fb, while i appreciate your work you provide your work here so you should provide all relevant info here aswell. it says hands in the name and some people come for exactly that but theres no information about the hands part whatsoever, no example pics with prompts or xy plots or trigger words. eventho i downloaded and tried for myself i can see the difference in skin but im not quiet sure if i see the effect on hands, so please go ahead and provide some info. especially with your pt situation right now, the offer might make people curious because hands are such a problem but they might not want to risk it just to try for themselves.
@JaysonPhoenix As I said: Try it out for yourself. :) I merged two LoRAs for male and female hands into it, but I guess, the KI is still struggling with this nearly "unlimited" human hand positions - such as every artist does. But I guess it "helps" a bit ... as I wrote in the description. Cheers!
PT format?into whitch folder
Models --> Lora // Jesus! :D // There is NO Malware in it. ^^
yes, i like it so far the most advanced skin lora, unless i missed something, thnaks a ton :-)
Cheers! <3
Might you consider uploading a safetensors version?
sigh :D It's already verified, you know. But yes: next version. :)
@Polyhedron_AI Cool waiting for next version.
@Polyhedron_AI thanks!
Everytime my system call for this Lora, I get this error... EVERY time, over the last day or so...
"Error loading embedding polyhedron_skinny_all.pt
Traceback (most recent call last):
File "C:\stable-diffusion-webui\modules\textual_inversion\textual_inversion.py", line 219, in load_from_dir
self.load_from_file(fullfn, fn)
File "C:\stable-diffusion-webui\modules\textual_inversion\textual_inversion.py", line 185, in load_from_file
assert len(data.keys()) == 1, 'embedding file has multiple terms in it'
AssertionError: embedding file has multiple terms in it"
it sorta messes up ALL my Ti's... of the many I have on my pc, only THREE show up now... since this error started.
@techken1 This is a LoRA, not an embedding or textual inversion. The correct folder is: models --> Lora // Put it there. :)
THIS is the file that's blowing up everything for me... "C:\stable-diffusion-webui\embeddings\polyhedron_skinny_all.pt" when it is called by any of the Lora's that it's called by.
@techken1 Put. This. File. Into. Lora. Folder. ;) It's NOT an embedding.
stable-diffusion-webui\models\Lora
@Polyhedron_AI Then. why. is. it. a .pt. file??? LOL
See image posted the gallery...
@techken1 Cause I screwed up. ^^ LoRA. Put. it. there. ;)
rename it? or just stuff it in Lora?
@techken1 No, it's working fine ... just put it into "lora"
@Polyhedron_AI K, stuffed, gonna reload, cause I'm getting errors everywhere from what ever happened. lol
@techken1 Make it so. :D
@Polyhedron_AI That did it, and all those nasty klingons are gone as well... ROFL
@techken1 Good! Earl Grey. Hot. :D
image post deleted
I don't know about that last part, seems Earl is getting a trifle old, and cooling off... How about a nice hot Irish coffee... with a double shot?
@techken1 Picard would prefer a red wine, I guess ... ^^
I'm talking about me, wanna join in? Makes for a great start to my day, getting ready to whip up another one.... The one from breakfast is getting thin in my blood stream. lol
@techken1 @techken1 I'm good. Thanks. ^^
@Polyhedron_AI Please upload a SAFETENSOR version of this LORA. PT files are too dangerous and it really looks shady to not have this as an option. It will create more downloads and traffic to this LORA if you do this.
@clevnumb If you find out how to convert this existing *.pt file into safe tensors without re-training I would do it: It's a merge of FOUR different custom LoRAs. I could not simply do it completely new, I'd would change. :)
@Polyhedron_AI Sorry I don't know how to accomplish this...
I'll wait for a safetensors version
Why are you wait for a safetensors version?
@Jinouga Pickled versions can run arbitrary code (aka hack stuff), safetensors cannot. It's best practice (read: smart and safe) to only use safetensors.
@clevnumb Oh, thanks! I didn't know that at all! It's important to know about cybersecurity best practices, so thank you! I personally have a machine on a Google Colab server, so I don't have anything to worry about lol.
@Jinouga other point is that formats other than safetensors doesn't seem to work with Easy Diffusion, it simply ignores .pt files until where I know.
what about pickle scanner?
@Natotela Pickle scanner is great if you need to double check one, but if you've already ran a compromised model...better to just use safetensors.
@clevnumb this! theres technically nothing stopping already running malware from injecting itself into your local pickle scanner forcing it to flag all other mailicious files as "safe". if you were breached once, someone could easily destroy your safety net from within.
also just bc safetensor is much safer, doesnt mean its safe either... better to just use safetensors from creators who seem like decent human beings
Why no safetensors?
Same answer here: If you find out how to convert this existing *.pt file into safe tensors without re-training I would do it: It's a merge of FOUR different custom LoRAs. I could not simply do it completely new, I'd would change. :)
@Polyhedron_AI It's a 5 minute job in Kohya's webui. Just use "Resize Lora" on the utilities page and give it the same size as your original.
@KGN I have as training parameters:
Mixed precision: bf16
Save precision: bf16
Network Rank (Dimension): 128
Network Alpha: 128
Is the "Desired LoRA rank" in "Rezise LoRA": 128 then? And "Save precision": bf16? Asking for a friend ... ^^ Thanks. :)
Safetensors please.
Same answer here: If you find out how to convert this existing *.pt file into safe tensors without re-training I would do it: It's a merge of FOUR different custom LoRAs. I could not simply do it completely new, I'd would change. :)
@Polyhedron_AI appreciate your work.
@Subte Thanks so much. :)
Would really like to try this, but I rather wait for safetensors version than to put a .pt file in my system. :/
Same answer here: If you find out how to convert this existing *.pt file into safe tensors without re-training I would do it: It's a merge of FOUR different custom LoRAs. I could not simply do it completely new, I'd would change. :)
Lol who cares. reeeeeee
Would prefer to do this in Koyha_ss; here are my training settings:
I have as training parameters:
Mixed precision: bf16
Save precision: bf16
Network Rank (Dimension): 128
Network Alpha: 128
Is the "Desired LoRA rank" in "Rezise LoRA": 128 then? And "Save precision": bf16? Thanks. :)
@Polyhedron_AI glad to hear that. Your work is awesome.
@Polyhedron_AI First off, thanks for your work, it's awesome! But if you get the chance, would you supply the official safetensors to this page? Or give someone permission to re-upload the safetensors version?
The reality is, ckpt/pt is "dangerous" and all these instructions for pt -> safetensors still require the one doing the conversion to load the "dangerous" model. Ie. If you load the pt model for conversion, you're still loading the "dangerous" part.
Although the suggestion here is a good idea if the .pynb is used on a remote machine like Google Colab to do the conversion, since it'll be Google's problem if there's something "dangerous" in it 😂
@rbmjay Thank you. :) Have fun with it. <3 // @conversion: I need someone who can explain me EXACTLY :D what to do without changing the LoRA itself. I can rezise it with Koyha_ss, but I need the exact parameters for that depending on my own training parameters (see above).
Where should I copy the safetensors file?
There currently is no safetensor for it. But anyway, all Lora's go into the '\models\Lora' folder.
@bonsai thanks
@bonsai Another question: Why is there nothing in the section for choosing the model in Control Net? Where should I download the models related to Control Net? Also, when I search for the base model used for these (lora) models, I cannot find anything.
What software is needed to do this? Can it be used for images already generated and saved as JPG?
I think you came here from google, this is for AI image generation. Yes you can apply it to already generated images.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.