ACM-VK | Fast SD Model

ACM-VK

Details

Download Files

About this version

The Flux variant. I feel like this is a lost cause. Despite very similar training settings, Flux seems to fight me on every front: characters, concepts, and anatomy. If you like skittles and pasties, Flux is just for you, but if not, you're better off with V1.

As usual, I tried training with standard tags on Flux, in order to not have to change how I generate images, while having the flexibility to allow for natural prompting. Does it work? That is yet to be seen.

Next time, I'll try raising the network rank and alpha to the same settings I do on pony-based models, to see if I could force Flux to actually learn new concepts and ditch the skittles/pasties already. (Default network rank and alpha in the training seemed ridiculously low, though there's the benefit of a small file size.)

Model description

This is what happens when I combine VHSPH with pov kisses. Tried to make a Flux variant of it, but from what I can see, it fights me on the concepts very hard, and there's lots of pasties/skittles despite my efforts--though to be fair, I set it to close to default.

In theory, this should do everything VHSPH does, with the added bonus of pov kisses, though if you plan to use the pov kiss stuff, it might look better on a square resolution. It can work on portrait sometimes, but you'd be better off adding "multiple views" when using portrait aspect ratios.

Images made by this model

Sort by

No Images Found.

Model Type	LORA
Base Model	Flux.1 D
Published	10/28/2024
Trained Words	Zhezhi Choco-chan, black hair, aqua hair, colored inner hair, glasses, mole Moca-chan, orange hair, orange eyes, pink eyes, gradient eyes Wankoin-chan, light brown hair, green eyes pov, viewer kiss pov, imminent viewer kiss pov, handjob, breast press, lying on person, imminent kiss