title: README
emoji: 📈
colorFrom: green
colorTo: purple
sdk: static
pinned: false
thumbnail: >-
https://cdn-uploads.huggingface.co/production/uploads/6631ebb1060ab1f6667d1f36/0R1T-WB8_VPItJJhnr5qo.png
Ars AI
Doing training and merging for shits and giggles
About
Ars AI is an independent, open-source generative AI playground driven by a very simple, highly sophisticated core philosophy: Random Bullshit, Go!
What started as pure curiosity about model merging quickly spiraled into an absolute obsession with pushing the limits of Stable Diffusion. We don't really do "orderly" or "cookie-cutter" standalone checkpoints here. Instead, every single release is a chaotic living experiment—a mad-scientist mix of aggressive UNet block merges, heavily stacked passive LoRA bakes, cyclical training loops, and sheer trial by fire.
Ars AI is the culmination of that wild journey. It's a space where we throw everything at the wall, analyze the wreckage, and turn every successful bake, every fried pixel, and every glorious "what if?" into a stepping stone for the next generation of models.
Philosophy
The "Random Bullshit, Go!" Doctrine
Knowledge isn't defined by what's safely tucked away in a textbook—it's defined by the absolute chaos of what we haven't broken yet.
- Every model is a reckless experiment. If we aren't throwing incompatible weights into the blender just to see what happens, we aren't doing it right.
- Every experiment generates raw data. Even if that data is just a horrifying, abstract wall of digital noise.
- Every catastrophic failure teaches us something. Usually, it teaches us exactly how many custom LoRAs it takes to completely fry a latent space.
- Every successful merge reveals another question. Specifically: "How did that actually work, and can we push it twice as hard next time?"
- Nothing is discarded. Even the most unhinged, ridiculous prompts and the most deeply cursed, unexpected results are meticulously documented in the lab notes.
FAQ
How many chaos makers are running the show at Ars AI?
There used to be a couple of others, but then life happened. So currently, you're looking at a grand total of one solo chaos maker steering this entire ship into the latent unknown.
Do you have a chaotic team backing you up?
Team? Not yet. As of right now, Ars AI is entirely powered by one overly curious hobbyist, a single MSI RTX 3060 Gaming X 12GB fighting for its absolute life in the rendering trenches, and an aggressively unhealthy amount of coffee.
(Yes, I used to have an ASUS ROG Strix 3090, but we're back in the trenches now, proving the "Random Bullshit, Go!" philosophy works on a 12GB card.)
What motivates your 'research'?
Officially:
We're throwing models into a blender and running cyclical training loops strictly for the shits and giggles.
(And honestly, seeing if we can get a 12GB VRAM card to punch in the heavyweight class is half the fun.)
Somehow... it keeps producing decent models.
Acknowledgments & Thanks
- Someone Very Significant
- OnomaAIResearch
- Cagliostro Labs
- nukeai1106
- GoofyAI
- Raelina
- DaoOwOarts
- Kohya_ss
- SeaArt AI
- PixAI
- Moescape AI
- Civitai
- Comte
- Minthybasis
- WAI0731
- And the entire generative AI community!
"Beyond the Visible."
There is no final model. Only the next experiment.