This is my first attempt at doing a model merge and easy is not right word for the experience. I tried to create a model that is not too focused on idealized figures for guys. You can get relatively ordinary men without much effort. The main keyword to use is 'abs' in the negative prompt. It should constantly produce results that are faithful to the keyword. Although it's not necessary, you can add the standard quality prompts that you see in every other image such as masterpiece, best quality, etc. The model can produce decent results without them. I haven't use negative embeddings on this model, so I can't say whether or not they will have bad impact on the image generation. Use the vae-ft-mse-840000-ema-pruned encoder to get aesthetically better results.
Here are the models that are used in this merge:
Checkpoints
Loras
This model is quite hairy and horny and the men may not often be ideally thin for some folks. Although the model can generate men of various ethnicities, it can be a hit or miss with those that have darker skin tones especially if you prompt any of the body types that indicates hairiness. I haven't tested it on women, but they should be rendered normally. I can't say for sure about nsfw generations. So far those the main flaws that I found with model. There might be more that I haven't caught but I'll try to improve whatever glaring flaws. I recommend using the resources after this section to fine tune or enhance your generations further.
So guys, enjoy the model!
Description
Initial version
FAQ
Comments (11)
This seems amazing, and by the prompts, intuitive to use. I need to test it <3
Thanks. I want it to be flexible and allow user to have some control over the gens.
I do agree :) nice job @BlackBear31 !
Thanks for including OhMen Lifelike in your merge 🤩 Keep up the good work man !
Btw, the model got a major update 2-ish weeks ago
Thanks. I had a hard time finding the right kind of models to merge because one of the models has a license that I feel was too restrictive at the time when I've done it. So I've looked for another one and tested a merge with your model, and that's how Regmen came to be.
@BlackBear31 My thanks to you man , hope you're having a nice WE :)
Can we talk on discord ? I'm j_wolf_
I am so bummed! I just downloaded this tool today, and it looks tremendous but all of my images are coming out very faded. Nothing like your gorgeous ones. What am I doing wrong? Thanks so much.
Use the vae-ft-mse-840000-ema-pruned file. It's a variation autoencoder, a file that suppose to assist the model during image generation. The file can be activated in Automatic1111 through an extension but I haven't used any other UI, so I don't had much of a clue on how to include it in the other ones. I had also realized that when I was testing the model but forgot to bake it in. I'll probably merge it into the model. The file should be in the description. I'm sorry about that.
Despite being someone who barely understands how this stuff works and trips his way through every model he finds on citivai, I've generated some really great images with this checkpoint. But looking forward to using 2.0 from here!
Thanks!