This is an early version!

When IllustXL gets to a more mature stage and the best practices for training and prompting are more clearly defined, this model will then be updated to a full release.

• Changed from sd1.5 to base IllustriousXL

• Improved dataset/tagging should make the model more flexible and better quality

• New version should be more receptive of different art styles while also able to keep the original style if desired

• New prompting structure with extra key words to accommodate model flexibility

• Trigger words: girlyumu, black hair, short hair,brown eyes,

• Optional additional words: narrowed eyes, terasu mc

• Additional negative trigger/key words: sketch, flat color, rough art,

• Weight: 1

• Trained on Illustrious-XL but I'm prompting it on WAI-NSFW-illustrious-SDXL

• Illustrious-XL works best with Euler a.

• Clip Skip 1.

• For resolutions lower than 1280x1280 Hires Fix is recommended for full body shots.

My personal preference for prompting her:

• main prompt:

masterpiece,best quality,amazing quality, girlyumu, black hair, short hair,brown eyes,

• negatives:

bad quality, worst quality, worst detail, flat color, multiple girls, sketch, anime coloring, simple background, rough art

Greetings,

I'm back from my hiatus feeling refreshed so expect to see way more content from now on.

To celebrate, I've updated Ayumu, Itsuki and Yuj to IllustriousXL.

I have a lot more stuff planned so I encourage you to join my discord which is where I post updates on what I'm currently working on

Here's my socials in case you want to keep in touch with what I'm up to: Discord X.com Pixiv

If you want to support me: Patreon Ko-Fi

Cheers.

//////////////////////////////////////////////////////////////////////////////////////////

Old stuff, please ignore.

Greetings,

Here's my take on one of TerasuMC's most infamous work. Although I'm personally not a huge fan of *that kind of plot* I can't deny that I absolutely love the design of the girls! And after many bad attempts and difficulties of working with a really limited dataset, I believe that the final result for most characters is satisfactory.

Also check out the other characters of this pack!

Model Characteristics

There are 4 different LoRA's available, one for each of the characters and each with its own trigger word.

• Trigger words: ayumu_\(terasu_mc\)

• Important: To avoid overfitting I recommend using the extension sd-webui-lora-block-weight and setting the weight to: OUTALL

• Use weight of 0.7 to 1. See what works best for you.

• I'm using DivineEleganceMix and Cartunafied and got good results, so I'd recommend using checkpoints that have an anime style.

• Play around with the samplers, but I got best results with DPM++ 2M Karras and DPM++ SDE Karras.

• Clip Skip 2 works best.

• The Ayumu LoRA can be a bit tricky to get good results due to limited dataset.

• Hires Fix works like a charm, I recommend using Lanczos or R-ESRGAN 4x+ Anime6B as the upscaler.

• Although not fully tested, I got some interesting results when using dramaticgecko1600 TerasuMC Art style, so I'd recommend to give it a try and see what results you get ;)

Training

I'm relatively new to LoRA trainning so any tips for improvement are much appreciated!

• Dataset size: 80 (Combined)

• Model: AnyLoRA

• Buckets enabled

• Epoch: 1

• Repeats: Good dataset - 100 & Bad dataset - 50

• Total steps: 12000 (Combined), Around 3k each.

I'd be happy if you guys could give me any character suggestions for my next LoRA!

Also, thanks for the 1k+ downloads!

Like my stuff? Buy me a Monster!

All my models and images will always be 100% free, don't feel obligated to donate!