Sign In

PhotoBoopV4

9
142
3
Updated: Jun 6, 2024
base model
Verified:
SafeTensor
Type
Checkpoint Merge
Stats
142
0
Reviews
Published
Jun 6, 2024
Base Model
SD 1.5
Hash
AutoV2
59C75D8F03

PhotoBoopV4 is the 4th iteration/attempt/experiment to create photorealistic image generation that I was happy with. Not that I was unhappy with some other models out there, there's just that need to make something that's "yours," you know?

Anyhoo, this was merged a while ago, and I wasn't particularly good about remembering what was used with what merge. This was based on ~boopv3 x [some photo realism model] which was in turn based on ~boopv2 x [another photorealism model] which, as a matter of course, was a result of ~boop x [a photo model]. The original ~boop is kind of a mystery to me since it was when I was first experimenting with merges.

For what it's worth, it mostly listens to prompts when it comes to clothes, though exact color matches are hit or miss. It will do SFW & NSFW and give you realistic results should you require various and sundry liquids. It seems to really only like 1 person at a time, though with prompt tweaking couples, and even the occasional threesome, can generate properly. It doesn't, however, seem to care too deeply about poses as often as I would like. I usually have to give it a +1.1 - 2.0 bump, depending on the pose in question.

PhotoBoopV4 will also completely misinterpret things or go off the rails with blank prompting. I look at it as relatively harmless as I'm generally just creating images for my own entertainment, but I understand that some folks wouldn't necessarily want their prompted woman lora to generate out as a glade with a path through it at dusk. I get it. I've posted a number of those kinds, actually, so I guess I do find them humorous...like the TI I'm trying to stop generating decorated plates of food despite not having been trained on plates of food.