theloraprodigy presents
Pokies LoRA
LoRA trained with theloraprodigy's high quality captioned dataset.
Use 2:3 ratio for best results (e.g. 512x768).
❤️Like this model? Support my work by buying me a coffee ☕
Commissions also available!
How to use this LoRA model? (AUTOMATIC1111)
Drop the LoRA into the "stable-diffusion-webui/models/Lora" folder and start AUTOMATIC1111.
Write your prompt as usual in e.g. text2img or img2img but also include "with pokies" when describing the clothing.
Example: "tanktop with pokies".
On the right, under the "Generate" button, click on the button in the center that has an image icon and that says "Show / Hide Extra Networks" when hovering with the mouse on it.
A new section under the prompt will appear, go to the Lora tab there and select this model
Training models requires expensive hardware and is consuming a lot of time. All of my models are free. You can support my work by buying me a coffee ☕
Description
Improved dataset
Added smaller boobs to dataset and removed cropped images.
FAQ
Comments (19)
Please check your v1.1 Lora, because I loaded it into Additional Network and I get this error when I try generate az image:
"NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check."
In principle, I use almost only fp16 models because of space constraints and I don't need high precision.
I wonder what is causing the error and the latest updates are up.
I just read that this was a problem in Civitas and the models were not downloading properly. I have now downloaded again and everything is fine.
Why does this not work? but some people can get it to work I tried both I am so confused.
Can you share your prompt and example outputs? I also just added a guide in the description.
it wont even generate while the first 2 are selected and tells me to put in launch augments to bypass errors. With the 2 launch augments it generates black images no matter the prompts. Im going to try your new one now tho
@theloraprodigy I downloaded your new update and it worked that other person might be right that it is just civitai
is it just me or are there like 3-4 versions of this lora with the same name but different authors?
This LoRA is trained from scratch. It uses regularisation images and captions, hence it is compatible with other LoRA's and won't be biased towards specific persons, age, clothing etc.
@theloraprodigy ok, but still,
https://civitai.com/?query=pokies
@dustystu One issue that is already happening is people are allowed to make terrible content and upload it to civitai.
I'm not saying that's the case for the different pokies stuff, but imagine someone makes the first Lora of "whatever you like." And it's awful. Terrible dataset, overfitted, etc. Should everyone just say "well dude x tried it first it's off limits now"?
There's a search bar and reviews. As this gets more popular hopefully the review system will work and you'll be able to log into civitai, search for pokies Lora, and then choose from 50 pokies loras. Then you can choose the one with 3200 positive reviews. And not the shit ones.
I want 50 more people to try this same exact concept.
@Jacksonhayle touche
(inb4 the sorting gets even worse and the site is overwhelmed with 5932 shitty jrpg waifu and Korean doll face Loras an hour. )
@dustystu True. Civitai isn't ready yet. I'm hoping that can be avoided.
failed to download v1.2..
We can’t connect to the server at civitai-delivery-worker-prod-2023-03-30.5ac0637cfd0766c97916cefa3764fbdf.r2.cloudflarestorage.com.
why are your picture prompts all fucked up? thats where i usually get my lora names from.
How many images in the training set?
144
I so need this in sdxl 😭
This doesn't work for sdxl :(
please can you do this in sdxl? i tried but no good results. thank you!
Details
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.