Naughty to Nice XL Slider

Differences from SD1.5

There are quite a few differences from the SD1.5 version. First, negative and positive are flipped. Yes, I wish they weren't. It's just the way things have to be. When trying to train it in the other order, results just end up worse. I've tried many times, I spent hours trying to flip the values also, with no luck. If someone can show me a functional way to do so, I'll put up a 1.1 with them flipped.

Second, There are a few differences in the exact results. Mainly, that SDXL has some differences in prompt execution resulting in the outputs being different, meaning the methodology had to be retooled heavily. I'm still tinkering with it, but it was nearly the same as starting from scratch.

Usage

-5 (Nice) to 5 (Naughty)

The biggest value to this LoRA (in my opinion) is that it bakes in around 50 tokens worth of data into the LoRA, and only uses single digits.

Obviously, use it to adjust characters to be more or less lewd. The XL version has a lot of differences from the SD1.5 version. I'm working on consolidating those differences, but it takes time, and XL takes much longer to train than SD1.5.

Lets be honest, 80% of use cases, you're probably using this in the naughty range. Why THINK about how to make a character lewd, when you're going to delete the image afterwards anyway? Just crank that slider up into the upper ranges.

The other 20% well, say you've got an image or an idea, but you're worried it might be too risque for what you're intended purpose is. Move that slider negative, and wipe away those features!

Prompt Advice

You may need to emphasize clothing in some prompts to keep the clothing theme!

Slider works best between -5 and 5. The range can go wider and produce viable images, as can the positive, but it gets quite lossy. Use non-anime models at your own risk. This was trained on anime, so consider that a warning as results can vary.

History XL

V1 - Methodology began at SD1.5, and went through dozens of iterations before a good adaptation for SDXL was found. V1 represents the first fairly viable model. I may work to get it even better in the future.

Training History

V1 - Trained using Sliders for Windows over 2500 iterations on aamXL, then range adjusted using Ostris's AI Toolkit. Unfortunately, results are far worse when trained to have the naughty side in the negative, so from here on they're the positive result.