Ok, so what you posted I think is the accelerate launch stuff, I have never used this, I am googling around and I think you execute a powershell script, is this done before or after the gui.bat or do you put in there. If you could run me through how to do this I would appreciate it. I am on Windows 11, so help in this department would be cool.

Thanks, another question,I know you have used hyperpreg, but I notice the Lora allows for multiple concepts. What do you use for class and tokens? What about the same class and token for both, like pregnant and pregnant vs sks pregnant for example. Also I heard class person is better than woman. But what about a class pregnant, or "preggo' for example

I'm training on ubuntu so if you want to use powershell, you would need to convert the script to something similar. I don't use the gui for training, just that script. Highly recommend training with WSL2 on windows, since training on windows alone sucks.

As for keywords/concepts, I just use the single keyword "hyperpreg" for that model. But you can train multiple concepts at once if you split your dataset into multiple folders. Personally I don't use classes or class images, too much extra work for me.

Anyway know a CoLab that can run Version 4?

Pretty sure A1111 has a colab version fork that could work

Bo could you provide a tutorial? Its a pain trying trying to find the right amount of steps, I'm also looking for some high res morphs for bigger bellies to train with. Anyone know of some? I was messing around with some vintage pregnant pics from imagefap, and a couple of low res morphs, and this is what I got. Lol. The lumpy ones actually make for great vore images. 

And yeah, these all have the usual negative prompts, bad anatomy, missing fingers, etc.

I'd love them to be more round though. Trained at 1400 steps, 400 classifier images. I mixed it with F222, might try hyperpeg next.


                       

Do you take commissions/challenges? I've got some pics I'd LOVE to see impregnated

Made these in Stable Diffusion A1111 webui, with epiCRealism as the basemodel, using Hyperfusion (the updated version of the hyperpreg ai model) as the Lora, with the following prompt, adjusting accordingly for hair color, location and stuff:

Positive- young, fit, beautiful woman, confident smile, nude, hyperpreg, roundbelly:1.5, gigantic belly, huge belly, veiny belly<lora:hyperfusion_279k_64dim-LoCon-v6:1>
Negative- cartoon, painting, illustration, (worst quality, low quality, normal quality:2) bad-hands-5 BadDream UnrealisticDream

Also used Hires fix with the 4x NMKD upscaler, and got them from 768x512 to 1536x1024. Steps at 20, CFG between 5-7.

They just need some inpainting and fixes but they came out pretty decent right out the bat imo besides the messed up hands.

(October 26, 2023, 6:36 pm)PregFan226 Made these in Stable Diffusion A1111 webui, with epiCRealism as the basemodel, using Hyperfusion (the updated version of the hyperpreg ai model) as the Lora, with the following prompt, adjusting accordingly for hair color, location and stuff:

Positive- young, fit, beautiful woman, confident smile, nude, hyperpreg, roundbelly:1.5, gigantic belly, huge belly, veiny belly<lora:hyperfusion_279k_64dim-LoCon-v6:1>
Negative- cartoon, painting, illustration, (worst quality, low quality, normal quality:2) bad-hands-5 BadDream UnrealisticDream

Also used Hires fix with the 4x NMKD upscaler, and got them from 768x512 to 1536x1024. Steps at 20, CFG between 5-7.

They just need some inpainting and fixes but they came out pretty decent right out the bat imo besides the messed up hands.

Excellent work!  Are you using this LORA? https://civitai.com/models/16928/hyperfu...80k-images
 I was under the impression it was used for Anime rather than realistic pictures.

Quote:Excellent work! Are you using this LORA? https://civitai.com/models/16928/hyperfu...80k-images
 I was under the impression it was used for Anime rather than realistic pictures.

Yep! Worked surprisingly well for realistic pics. Did not expect that. Honestly, HenmixReal is a better base model to use along with the hyperfusion LORA imo, tho, just gotta use different settings.