Hello Gay AI community.
https://discord.gg/A3RTYYkCHY
I am the creator of this model and I need your support to continue the project. This is a model crafted by labeling dozen thousands of images with hand-labeled data. About 2~ years of work.
This model excels at creating fantastic muscles and flaccid/erected dick.
It uses natural language as prompts so you should prompt a bit differently than usual. In some areas this model is smarter than other image AIs because it tries to understand language than tags.
It recognizes artist names Celery Man, Holo Eden, Silverjow etc
Use "Celery Man" for soft cozy vibe
"Holo Eden" for spicy colorful vibe
"Silverjow" for stylish detailed but a bit toned down colors
"PatLee" for intense photoshoot images
Make sure to start your image with "photo by PatLee" or "illustration by Artist"
Description
Further trained on dataset
FAQ
Comments (10)
Can you please post the prompt along with the images so that we can see how you get these stunning images. I am not able to get anything close
I highly recommend to join the discord channel as there are thousands of countless examples. Also as we are a community that's training the AI collaboratively, the styles of your prompting must change with each iteration cuz our AI is getting smarter each time, and I have no idea how to prompt better each time, as this is a process done outside my control.
as for your info, I prefer DPM2/3
use the "beyond negative embedding" from this site...its all you need...this checkpoint is awesome with that
Can you please include some basic information. ie.
Which Sampler works best. Which vae are you using. Steps, CFG etc...
Images with prompts are the quickest way for everyone to see how you achieved the images, even a quick note in the comments section with the prompt wording you used would be helpful.
Thank you
I have tried the model with a tone of Samplers and the images are nice but not to the quality of your samples
I highly recommend to join the discord channel as there are thousands of countless examples. Also as we are a community that's training the AI collaboratively, the styles of your prompting must change with each iteration cuz our AI is getting smarter each time, and I have no idea how to prompt better each time, as this is a process done outside my control.
as for your info, I prefer DPM2/3
@blu3bug Thank you for replying and for the sampler tip
https://civitai.com/models/108821/beyond-negative-embedding use this, its all you even need in the neg...youll get images like those above
I'm using:
DPM++ 3M SDE
Sampling steps: 20-25
Hires. fix: ESRGAN_4x; Hires steps: 0; Denoising: 0,7, Upscale: 2
Refiner: CFG Scale: 7
Let me just say, this isn't just for gay guys. 😏
This is a really great checkpoint, and it's giving very good results, but it sure loves to cut the picture in the middle of the hair. It's very often missing a few centimeters up and it's a pain to outpaint. Your own pictures have this problem as well. If it's something you can fix in future versions it'd be great.