Obsessive Compulsive Disorder v2

Alright guys, v2 is finally here. I’ve spent a lot of time tearing the original model apart and rebuilding it to improve the v1. If you liked the first version, you’re going to love this one, but heads up: it’s a whole different beast now.

Instead of just being "good at details," I’ve turned this version into a Latent Behavior Engine. That’s a fancy way of saying the model now actually understands the soul of the materials and styles you’re asking for.

The "key" that open all doors: devildonia_style

First things first: the v1 triggers are still here, but devildonia_style is now the MVP. Think of it as the glue that holds everything together. If you want the model to really "get" an artist's style or a specific look, keep this tag in your prompt. It adds a ton of stability and helps the model navigate all the crazy stuff it’s learned.

The other two (flawless_style and realistic_skin_texture) still work great, but honestly? Try using them one at a time when you really need that specific finish.

More Artists and Characters than I can count

This model is a massive merge, which means it’s packed with knowledge. I’m including a list of 250+ artists that I’ve personally tested and that look absolutely killer. They don’t just "filter" the image; they change the whole vibe.

As for characters... man, it knows a ton. Since it’s a merge, it’s impossible for me to list them all or test every single one, so I’m leaving that to you. Go explore, try your favorite characters, and see what happens. It handles Furry and Anthro stuff like a champ too, with way better anatomy than before.

My Advice: Stop over-prompting!

One thing I noticed: people love to drown their prompts in "quality" tags like 8k, hyperrealistic, cinematic lighting... Don’t do that here. You’ll just choke the artistic DNA of the model.

If you want it to look amazing, keep it simple. Start with this and then add your scene:

Positive: masterpiece, best quality, devildonia_style, [your stuff here] Negative: worst quality, low quality (and maybe your favorite negative embedding).

The Nerdy Stuff (What works for me)

The model is super flexible, but if you want that "chef’s kiss" artistic look, here’s my favorite setup:

• Sampler: Heun
• Scheduler: Beta
• Steps: 20 to 25
• CFG: I usually stay between 5 and 7 for stability, but this thing can go all the way up to 18 if you want to get really "obsessive" with the details.

If you just want a clean, standard image, Euler Ancient with the Simple scheduler is a solid choice, though it’s not as "artistic" as the Heun/Beta combo.

Compatibility & Showcase

For this v2, I’ve made the showcase images using Forge Neo. I wanted to show you guys that the model works like a charm not just on ComfyUI, but also on Forge and Automatic1111.

If you are a ComfyUI power user and want the specific workflow I usually run, you can just grab any image from the v1 version and extract it from there—it still works perfectly with this new version!

Go wild, push it too far, and please—share your masterpieces in the gallery. Seeing what you guys create is what keeps me going. If you like what I’m doing, drop a follow and a reaction. It’s free and it means the world to me. 🫶

— DeViLDoNia (A.K.A. HaDeS)