COMPASSMIX XL 

Model Download
The model is available from CivitAI here.

Check out fast inference of this model at frosting.ai, which is the primary sponsor of this model.

Recommended Settings
We recommend using the following settings with our model for best results.

  • Resolution: 1024x1024
  • Steps: 8-12
  • CFG Guidance: 2.5
  • Sampler: Euler or Euler a

The model utilises the low guidance and steps due to it being an SDXL Lightning model. It's advised to keep your step count as well as your guidance low, whilst you may increase the step count if you choose to do so, there is a clear point of diminishing returns, therefore we recommend you to follow our recommended values here.

Model Info
This model retained much of the knowledge from base SDXL, leaving an enormous array of concepts that the model understands while also understand a lot of furry concepts.

Strengths & Weaknesses
During our evaluations of the model by ourself and the community, we find the following strengths and weaknesses in the model.
Strengths

  • High-level natural language understanding with the ability to handle multiple concepts.
  • A good level of tag understanding, allowing for a mixture of prompt styling.
  • A wide range of art styles and following of art styles.
  • Great prompt coherence
  • Ability to do humans and anthros well, with a wide range of supported species.

Weaknesses

  • Noisy looking images, especially with photo backgrounds and plain backgrounds
  • Scenes with duo and higher tend to have various anatomical problems.
  • The model tends to prefer shorter captions compared to longer more detailed captions.


Prompting & Tips
Natural language usage is recommended as the model and SDXL in particular understands it. However, you are also welcome to use tagging as you please.
This model is based off of SeaArt, it's advised to use their quality tags. They provide an example of this from their HuggingFace page.
The model supports various levels of quality hints, from "masterpiece" to "worst quality". Be aware that "masterpiece" and "best quality" may lean towards nsfw content.
They provide other prompting tips on their HuggingFace page that may be applicable to the model, you can find that page below.

Credits
This model was developed by the Horizon Team, a team dedicated to creating high-quality furry AI models.

  • Lodestone Rock
  • Bananapuncakes
  • Thessalo


We also thank SeaArtLab for training the original SDXL furry model that was the basis of this merge.

Extra Funded Provided By:

Finally, we'd like to express our heartfelt gratitude to our wonderful Patrons who made this model possible:
You are awesome! Patron:

  • Nexus
  • Nothingburger1234
  • TheGreatSako

Serious Patron:

  • SunBee
  • Simon
  • AndriFox
  • Mobbun
  • Kadah
  • Zenko

Patron:

  • Joseph
  • Krazark
  • kcanis


You can also give a direct PayPal donation via our PayPal link here.

Keep in Touch
You can reach us at our Discord server!

Enjoy the release everyone! Take care.