High-resolution images: Trained on 1024x1024 images (default for XL).
Various art-styles: Multiple artists/illustrators from the 1970s - 1990s. A list of artists tags can be found bellow.
Easy to prompt: Stylization of simple prompts can be achieved with as little as one special token (trigger) appended to the end of your prompt.
True to source: Special attention was payed to insure that the artstyle(s) being generated are as close to the individuals artstyle as possible.
Flexible: Can be easily merged with other SDXL checkpoints / LoRAs for more expansive generations.
Full parameter fine-tune: This is a full parameter fine-tune of the original SDXL base 1.0 model. Both the U-Net and both text-encoders were tuned during training.

General Information:

Base model - SDXL v1.0
VAE - SDXL-vae-fp16-fix
Type - Full-parameter fine-tune, including text-encoders, bfloat16 (bf16).
Style - Painting, Drawling, Illustration, digital painting

Usage

Special Tokens:

BAStyle, a general prefix that is not strictly required but can help maintain the models style.
- Place at beginning of your prompt.

Artist Tags: Appending these tags to the end of your prompt will guide the model to generate an output in the style of the artist that's being tagged.

by Boris Vallejo
by Brom
by Caza
by Chris Foss
by Clyde Caldwell
by Col Price
by Dave Dorman
by Frank Frazetta
- sketches by Frank Frazetta
by Kelly Freas
by John Berkey
by Gimenez
by Ken Kelly
modern art by Kirsi Salonen
modern art by Kyu Yong Eom
by Luis Royo
by Michael Whelan
by Mike Hoffman
by Milo Manara
by Moebius
by Philippe Druillet
by Sanjulian
by Ted Nasmith

With more to come in the future!

Random Style Generation (RSG) Tag: The following token (trigger) is a made-up artist that generates an image in the style of a random artist. It's not perfect but can yield interesting results

by Gladius

Wildcard: A wildcard that includes the various artists can be found here.

Limitations

BAM! inherits some (not all) of the limitations from the original SDXL v1.0 model. Specific limitations:

Bias: This model is bias towards paintings, drawings, and illustrations from the 1970s - 1980s.

NSFW: NSFW generations, while possible, are limited. Improved when run alongside a LoRA or merged with another SDXL checkpoint.

Questions/Feedback/Updates?

Visit my thread on the Unstable Diffusion Discord

Future updates are already in the works to bring SDXL up to — and beyond SD1.5 standards. If you want to contribute, DM me at the discord thread linked above. My goal is to create fine-tunes on unique styles or concepts SDXL struggles with, expanding the pool of models for merging.