Type | |
Stats | 5,062 163,917 |
Reviews | |
Published | Jul 3, 2023 |
Base Model | |
Training | Steps: 2,500 Epochs: 5 |
Hash | AutoV2 9FB965D556 |
Here's a new version of my "bulgerk" series. This time I trained on images of real people with "dickprint" or bulges which are more than just a round bulge. The original "bulgerk-soft" is here (it's a textual inversion) but I find this one gives bigger (hornier) results.
Some tips.
I tried to keep the training "low" to avoid overburn.
Some models (protogen fore example) are more "intensive" meaning that a low CFG scale of 4 or 5 is desirable
While some other models like virile reality might work at CFG scale of 6 for example
But make sure to pair this with a weight in the LoRA:
<lora:bulgerk-dickprint:1> should work in most cases, but it's close to "burning".
<lora:bulgerk-dickprint:1.2> is pretty much the max I've seen without weird stuff coming in
<lora:bulgerk-dickprint:0.6> will make it less unrealistic but still nice.
So for example, if you do
<lora:bulgerk-dickprint:1.2> and a CFG Scale of 4, it might work.
Or reverse, <lora:bulgerk-dickprint:0.5> with CFG scale of 7 might also work.
Prompting: I didn't bake in any prompt in the LoRA. But to get it to show up, adding bulge, or crotch bulge early in your prompt helps a lot. "A man with a crotch bulge".
The types of pants also seems to affect the results a lot. Jeans, chinos, cargo pants, shorts, sweatpants, etc...