So what's new for REvalitize v0.3

So to start off with a quick vent, when making V 0.2 using Civitai Lora training was absolute hell on earth, ITS SO GODDAMN BUGGY AND AFTER HOURS AND HOURS OF FAILED AUTOTAG NONSENSE

...I gave up. So unfortunately it's not really a definite version but I'll still leave it here as something to play around with.

Ended up retraining my dataset again on another website and making sure the legs and hands were coming out ok for the lora as well as the background setting. The original V 0.1 had only been focused on indoors, so this one has more reach for other settings, this will still be improved though so look forward to that.

As said before this is a "Butt Focus" lora, but can also be used to pop out other aspects of the body, with some nice lighting. I trained this new dataset by using many different light sources to create some good looking shadows around the bum area. You can also use the lora for any explicit positions you like.

PROMPT STYLE:

Now this time it is highly recommended that if you want the same style as the photos try not to mess around to much with the negative prompt.

For example for most of my photos I simply put for NEG prompt...

"blurry background, artifacts, bad hands, loli, child, separated legs, missing limbs, fused limbs, fused hands"

And for positive you can put whatever you want. Most of the embeddings is already merged with my lora, so if you add it in the negative or positive prompts, if you want, it could potentially change the style of the photo. But still experiment!

Now for Lora strengths, samplers and combinations:

Keep REvalitize lora at around 0.6 to 0.8, going to 1 will have the character default to some nice looking lingerie with the same low angle pose, so if your looking for clothing variety, improve your prompts and lower the strength a little.

For a more 2.5D while also keeping the details of your character I would suggestion using the lora called Unknown slider, at -2.5 strength.

For some reason going negative makes the character less 2D and more 2.5D-ish. Going positive also makes the photo more better aesthetically with ALOT of variety but loses the 2.5D style, at least after some experimenting with it. This what I noticed on multiple models.

My samplers used are Euler_Ancestral/DPM 2M Karras: Karras, Beta, Beta57 for schedulers with simple Ksampler.

But like I said always experiment, these are just my defaults for the pictures above based on different models.

And speaking of models, you no longer have to just focus on using DREAMSICLE, please explore other models!

Anyways keep the buzz support for more loras! And remember have fun!

PAST NOTES FROM THIS POINT....

|

|

|

|

INTRODUCING REvalitize Project Lora:

This is super experimental, essentially makes the ass pop out more at a low angle, while adding more dark realistic shading, with shiny wet skin. Try it out. This is my first lora so results may vary.

Highly recommended to use this bare bones quality prompt in positive prompt to activate lora:

"ultra-detailed, ass focus, sexy woman, huge ass, amazing quality, very aesthetic, absurdres, newest, scenery, high detail, (dramatic lighting), detailed hands, very aesthetic."

Neg prompt (includes embeddings such as lazyhand and lazyneg)

"worst quality, low quality, blurry background, front view, 2d body, multiple people, embedding:lazyhand, embedding:lazyneg"

(if you haven't downloaded any embedding files, delete this part of your negative prompt but it's highly recommended you get it)

I recommend using Dreamsicle model, but any model you like is fine. I'm not sure how it will be with other models but hey always experiment!

More attempts will come through. Results will vary, has a tendency to produce multiple legs, bad hands for now. will be fixing in the next version after collecting more buzz.