Model Info
This model is intended to create antro/furry art, I primarily use it to create semirealistic art, but it can be used to create a wide variety of styles. This model uses both danbooru and e621 tags for prompting which can without lora generate characters, clothes, poses and more. This model uses tagging and is not proficient with natural language and has a baked in vae, so no seperate vae is needed.
Recommended Settings
Updated 2025-12-20
Resolution: Any SDXL resolutions, I use 1360 x 768
Steps: 25
CFG Guidance: 3
Sampler: Euler Ancestral
Scheduler: SGM_Uniform
It is also highly recommended to use Hires fix and optionally FaceDetailer (ComfyUI) or ADetailer (stable-diffusion-web-ui/A1111, stable-diffusion-web-ui-forge).
Recommended Prompts
Updated 2025-12-20
Positive
masterpiece, best quality, newest, absurd res, amazing quality, very aesthetic, very awa, anthro, furry, (no humans:0.1),Negative
worst quality, bad quality, human,
bad hands, deformed,
watermark,For further ideas and inspiration check out the prompts i used on the images associated with the model. They may not use the latest version of the prompts in their generation.
Changelog
V 7.0 Updated to Illustrious. Improved image fidelity, improved ability to make detailed poses.
V 6.0 Improved character anatomy, mitigated risk of human appearance, mitigated risk of names appearing.
V 5.0 Updated to latest version of source checkpoints as of 2025-06-11.
V 4.0 Updated to NoobAI. Improved character anatomy, color composition and prompt adherence.
V 3.0 This Illustrious based model has significantly improved versatility and offers the full spectrum of both danbooru tags and e621 tags.
V 2.0 A new hybrid merge between SDXL Lightning and Pony models. The model achieves significant improvements in anatomy, prompt adherence and overall quality while retaining the vast species library of the SDXL Lightning model.
V 1.1 Significant refinement of the 1.0 version including new recommendations in lora, generator settings and e621 artist prompt tags.
V 1.0 Initial upload.
Credits
Thanks to the SeaArtLab team, horizonteam, freek22, Urameshi, shadowrui, Jedas as well as Desync for creating and iterating the SDXL based models used in this merge.
Description
This Illustrious based model has significantly improved versatility and offers the full spectrum of both danbooru tags and e621 tags.
FAQ
Comments (7)
Yes, I'm the one who's been adding in all the plus sized artwork lately.
But what can I say though? The model pretty much does everything I need it to do; easily switch between stylized 2D artwork and realism without relying on heavy use of tags or LORAs, it can work with some of the existing LORAs I've been using for Pony models, and it pretty much doesn't run into any issues with raw prompting like I'd get with the other models I've been using before.
About the only complaint I have (although this seems to be more because it's Illustrious) is just the amount of quality and base negative tags you need for it to work. But once you "set it and forget it" it isn't really that much more of an issue.
I´m glad you enjoy it, you are correct that quite a few prompts are needed to produce the best result. However just as you said this is a "set it and forget it" situation.
I´m constantly working on iterating on the prompting and fine-tuning for the best result as well as searching for new and updated models to use in this merge.
I hope you will continue to enjoy the model and share your creations ╰(*°▽°*)╯
How to fix the random color artefacts? Sometimes the image has a detail with sharp color or there's a random color spot on the image, sometimes the color spot malforms the face - is there a fix for it?
What prompts are you using? What sampler are you using? What scheduler are you using? What cfg are you using? How many steps are you using? Are you using hi res fix?
@Pentriac The thing is, I am using HiresFix. But it's Stability Matrix's Inference (ComfyUI). First pass is Euler, Normal, 15, 3.0, denoise 1.0. But HiresFix is different: Euler Ancestral, Normal, 20, 7.0, denoise 0.7. Prompts are NSFW kind, nudity. Does this model require special handling to avoid the color artefacts? The artefacts, malformations due to color bomb, random color spots, sharply colored details occur 50% with the above settings.
@lyagushka420 The model does not require any specific prompting to avoid color issues. In my tests I have very rarely found such an issue and it has never remained after a hi res fix apply. I would strongly recommend using Euler Ancestral with SGM_Uniform for both your initial creation and the hi res fix pass. I would also recommend that you keep the same cfg on both the initial creation and the hi res fix pass, and that you lower the denoise on the hi res fix to between 0.35 - 0.5. I hope that helps, otherwise experimenting with different negative prompts may also address the issue.
@Pentriac Okay, I'll try that out. My negative is already half a letter long. But the console displays the warning that I exceed the 77 token limit.