BoltIL | IllustriousXL版本v1.0 (ID: 998798)

A stylized model for Illustrious. This model makes use of dataset of fantasm and some other pastel images I had, which failed on pony.

v1

The Model is trained/LoRA merged on PaSanctuary (SDXL, IllustriousXL) by @FallenIncursio. That also means it carries over the usage of tags anime | realistic | 3d, where anime or well nothing is supposed to be default style. BoltIL carries over the dark feel of the base model as well.

v2

This version is a lora merge, style blocks only (thanks @PotatCat for extracting them out for me) in NoobAI XL (NAI-XL) Epsilon-pred 1.0-version. It still reacts to tags like anime | realistic | 3d | nothing(that is adding no tag) as earlier. Also have effect when used in negative

Usage

Positive: masterpiece, best quality, high quality additionally add anime | realistic | 3d if you want

Negative: the usual negative that other IllustriousXL models use, a fuckin wall of text (thanks for having such shit captioning, next time hire people who know what they're doing like @Anzhc or @bluvoll) lowres, worst quality, low quality, bad anatomy, bad proportions, extra limbs, extra digit, extra legs, extra legs and arms, disfigured, missing arms, too many fingers, fused fingers, missing fingers, unclear eyes, username, simple background

Vae: Vae is baked in. For custom Vaes, use anything trained by @Anzhc (https://civitai.com/models/888145/aaa-anime-sdxl-vae or https://civitai.com/models/855441/aaah-vibrant-vae-i-guess), the lower memory one https://civitai.com/models/140686/fix-fp16-errors-sdxl-lower-memory-use-sdxl-vae-fp16-fix-by-madebyollin or the base sdxl

Make use of adetailer for face, BUT, most of the time the eyes will look fucked. Reason? Well, that's pastel for ya.

Recommended adetailer models: Again, any adetailer model trained by @Anzhc instead of stuff that comes with adetailer will work (https://civitai.com/models/293448/anzhcs-face-segmentation-prototype-or-yolov8-or-adetailer-model)

Sampler: Euler A or DPM++ 2M

CFG: 5-7

Steps: 20-30

Base resolution: 1024

This model is supposed to work with resources trained on Illustrious-XL-v0.1. Other SDXL or pony resources may not work.

Now for the name choice I have.... well the story goes something like this.....

props to @MadlyLaughing for the name.

Cons/Issues

• Can sometimes fuck up with images, making them too cluttered. The reason for this is I hate the washed out images that base models give, and then there is this weird behavior where clothes would come out like burnt and hair will have a mix of realism in it, so I merged the LoRAs at full strength in a hope to reduce that, but that have a side effect as well.

• skin colors especially dark/black can get patches of light skin like below bangs and face, idk why it happens.

Recipe: PaSanctuary + ftb.safetensors(weight 1) + mmmmmm.safetensors(weight 1)

This Model is under Fair-AI-Public-License-1.0-SD. Check out the description on original base(Illustrious-XL)

The LoRAs used in this model are linked in the download section for anyone to view.

描述:

训练词语: anime,realistic,3d,no triggers and no this is not a trigger, this is to tell people you dont necessarily need to use a trigger cuz people on civitai lack skill called reading