<hr /><hr /><h3 id="obsessive-compulsive-disorder-v0.1-(unreleased-base)">Obsessive Compulsive Disorder v0.1 (Unreleased Base)</h3>Stumbled across the original v0.1 checkpoint while clearing out drive space today. It was never published at the time, but since it's the actual foundation behind the whole OCD series, I didn't want to just leave it sitting around collecting digital dust.It's an earlier, rougher experiment, but for anyone curious about the model's roots, here it is.I haven't included recommended settings, so just check the showcase images for generation details, or feel free to grab the experimental ComfyUI workflow included if you want to use the exact pipeline I used.Have fun with it!HaDeS<hr /><hr /><h2 id="obsessive-compulsive-disorder-v3-gamma-update">Obsessive Compulsive Disorder v3 - Gamma Update</h2>Gamma Visuals: Drastic improvement in color balance and shadow depth. No more "burnt" or oversaturated looks at high CFG.<h1 id="obsessive-compulsive-disorder-v2">Obsessive Compulsive Disorder v2</h1>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.<h3 id="the-&quot;key&quot;-that-open-all-doors:-devildonia_style">The "key" that open all doors: devildonia_style</h3>First things first: the v1 triggers are still here, but <code>devildonia_style</code> 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 (<code>flawless_style</code> and <code>realistic_skin_texture</code>) still work great, but honestly? Try using them one at a time when you really need that specific finish.<h3 id="more-artists-and-characters-than-i-can-count">More Artists and Characters than I can count</h3>This model is a massive merge, which means it’s packed with knowledge. I’m including a list of <a target="_blank" rel="ugc" href="https://drive.google.com/file/d/1lT-AoU7WzzV373L3tr170fcvf6XpbD7P/view?usp=drive_link">250+ artists</a> 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.<h3 id="my-advice:-stop-over-prompting!">My Advice: Stop over-prompting!</h3>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: <code>masterpiece, best quality, devildonia_style, [your stuff here]</code> Negative: <code>worst quality, low quality</code> (and maybe your favorite negative embedding).<h3 id="the-nerdy-stuff-(what-works-for-me)">The Nerdy Stuff (What works for me)</h3>The model is super flexible, but if you want that "chef’s kiss" artistic look, here’s my favorite setup:<ul><li>Sampler: <code>Heun</code></li><li>Scheduler: <code>Beta</code></li><li>Steps: <code>20 to 25</code></li><li>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.</li></ul>If you just want a clean, standard image, <code>Euler Ancient</code> with the <code>Simple</code> scheduler is a solid choice, though it’s not as "artistic" as the Heun/Beta combo.<h3 id="compatibility-and-showcase">Compatibility &amp; Showcase</h3>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!<hr />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)<hr /><hr /><h2 id="obsessive-compulsive-disorder">Obsessive Compulsive Disorder</h2>Obsessive Compulsive NSFW Disorder is a stylistically flexible NSFW model designed around controlled obsession with form, detail, and erotic aesthetics.The model is merged on a large and diverse dataset, providing strong generalization, high consistency, and excellent detail retention across a wide range of NSFW scenarios. It also performs exceptionally well with furry-based imagery, handling anthropomorphic anatomy, textures, and proportions with notable accuracy and stability.In its natural state (no triggers), the model delivers a balanced, raw output that reflects its core training—clean composition, strong anatomy, and expressive sensuality.The model includes optional style triggers that allow you to push the visuals in specific directions:<ul><li><code>devildonia_style</code> Activates a distinctly European erotic aesthetic, inspired by iconic 20th-century erotic illustrators. Expect elegant lines, mature sensuality, and a classic artistic feel.</li><li><code>flawless_style</code> Shifts the output toward a hybrid look with stronger lineart influence, cleaner contours, and a more stylized, controlled finish.</li><li><code>realistic_skin_texture</code> Enhances skin depth and micro-volume, adding a more tangible sense of surface and subtle realism to the image.</li><li></li></ul>These styles can be combined, but best results are achieved by adjusting trigger weights rather than stacking everything at full strength. Experimentation is encouraged.All showcase images include an embedded ComfyUI workflow, allowing you to inspect, reuse, and adapt the exact pipeline used to extract the maximum quality and consistency from the model.Enjoy pushing it too far—carefully.Give me all your love, follow me, react, post your masterpieces... It's free and it helps a lot.— DeViLDoNia (A.K.A. HaDeS)<hr /><hr /><h2 id="v1.01-dmd2">v1.01 DMD2</h2>Minor improvements and bug fixes. Merged with a custom DMD2 LoRa, as the checkpoint perfectly adapts to its characteristics without compromising quality or detail.<h3 id="sampler:-euler-a">Sampler: Euler A</h3><h3 id="scheduler:-simple">Scheduler: Simple</h3><h3 id="steps:-12">Steps: 12</h3><h3 id="cfg:-1.4-2">CFG: 1.4 - 2</h3>

Obsessive Compulsive Disorder

obsessiveCompulsive_v01.json

obsessiveCompulsive_v01.safetensors

Files with SHA256: 80219089a41982c0288d3728502c4aa5cb89f73c651de98d2033b0f16390d9d4