👋 If you like what I do and want to support the development, feel free to buy me a coffee:<a target="_blank" rel="ugc" href="https://ko-fi.com/herrscheragga"><img src="https://camo.githubusercontent.com/7765e3dbce77c82639ff14d27dda26628b206727b61a6736190f90e81c78a121/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4b6f2d2d66692d537570706f72742532304d652d4646354535423f7374796c653d666f722d7468652d6261646765266c6f676f3d6b6f2d6669266c6f676f436f6c6f723d7768697465" alt="Ko-Fi" /></a><hr /><h3 id="neural-repair-and-portable-checkpoints-lora-type.">Neural Repair &amp; Portable Checkpoints Lora Type.</h3>Hello! I'm back with something much juicier than ever!Originally, I planned to release more Samplers (and I will), but I pivoted to solve a critical flaw I found in the community: Many popular merged checkpoints have a corrupted Layer 11.<ul><li>This translates to:<ul><li>❌ Text Encoder errors (NaNs).</li><li>❌ Poor LoRA compatibility.</li><li>❌ Massive information loss.</li></ul></li></ul><ul><li>And here's the solution: [<a target="_blank" rel="ugc" href="https://github.com/HerrscherAGGA/AGGA-STL/tree/main">Anti-Nans + RAM Cleaner</a>] Uh-huh... The LoRA repair method failed, so I engineered a Runtime Fixer. Just paste the provided script into a new cell in your Colab/Notebook, run it before the WebUI, and you are good to go.<ul><li>Herrscher Shield: Scans Layer 11 and eliminates NaNs in RAM instantly. No need to download fixed checkpoints.</li><li>AGGA Optimizer: Aggressively cleans RAM to prevent Colab crashes.</li></ul></li></ul><hr />Okay, now let's talk about these Loras (Total / Duplo / Ultra):These LoRAs function as "Structural Converters."Think of them as a high-end "Cosplay" for your checkpoints: they allow a lightweight model to adopt the exact visual DNA, intelligence, and stylistic precision of a massive 6GB checkpoint (like Pony or Illustrious).Instead of dealing with architectural instability or NaNs, I have distilled the core features of these giant models into optimized 500MB-1GB files. They let you inject the prompt-understanding and "soul" of a heavy base model into any other refined checkpoint without the overhead of downloading or loading 6GB files every time.<ul><li>In conclusion: Now you have a tool for every need:<ul><li>Do you just want a refined DMD? -&gt; Total.</li><li>Do you want the information and style? -&gt; Duplo.</li><li>Do you want it all? It'll be a Copy -&gt; Ultra.</li></ul></li></ul><hr />🟢 TOTAL (Concept Injector):<ul><li>What it extracts?: Text Encoders + <code>attn2</code> (Cross-Attention).<ul><li>Complete Version: Extracts Text Encoders + attn2 (Cross-Attention). It’s the "Brain" of the model.</li><li>Visual Only Version: Extracts only the Style (UNet). | DMD2 Pure.</li></ul></li></ul><ul><li>What does it do?: Associates words with concepts.<ul><li>Total knows that "Miku" means "Teal hair, long pigtails".</li></ul></li><li>Result: Corrects what is drawn, but the "brushstroke" remains from your base model.</li></ul>🟡 DUPLO (Structure &amp; Geometry):<ul><li>What it extracts?: Text Encoders + <code>attn2</code> + <code>attn1</code> (Self-Attention).</li><li>What does it do?: Controls geometry and spatial composition. <code>attn1</code> is where the "shape" of the style resides (eye size, body proportions, composition).</li><li>Result: The image gets the structure of the source model (e.g., Pony), but the rendering (skin, lighting) is a hybrid.<ul><li>Best for fixing anatomy while keeping your checkpoint's texture.</li></ul></li></ul>🔴 ULTRA (Full Replica):<ul><li>What it extracts?: EVERYTHING (<code>attn</code>, <code>ff</code>, <code>proj</code>, <code>te</code>).</li><li>What does it do?: Copies the FeedForward (FF) layers too, which determine the Render Style (lighting, line weight, shading).</li><li>Result: A complete conversion. The base model visually disappears and becomes a perfect replica of the source.</li></ul><hr /><h3 id="important-versions-and-warnings">⚠️ IMPORTANT VERSIONS &amp; WARNINGS</h3>🟡 Visual Only vs. Complete (Zip)<ul><li>Visual (Online Gen Friendly): Use this for quick style transfer.</li><li>Complete (Zip): Includes the "Fixed" files that connect text properly. Use this for serious work.</li><li>Note: I fixed Duplo, but the IL &amp; NoobAI base is sensitive. Treat it with care!</li></ul><hr />⚠️ Visual Only Usage Note: Don't be scared! Even though this is based on my DMD2 architecture:It works perfectly at HIGH STEPS (20-30+) without burning (great for detailing).It works perfectly at LOW STEPS (4-8) for speed.Wink wink 😉<hr /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/3af09aca-fd27-4534-95a0-11638a5f5cfd/original=true/3af09aca-fd27-4534-95a0-11638a5f5cfd.jpeg" />Thanks so much for your support! ♥

DMD2: Smart-Brain Infusion (The Portable Checkpoint Series)

Hi, I was asked for a DMD that's compatible with TF, I'm not sure if I managed it, those who are into that can let me know. This brain includes quite a bit of furry information, although I also reduced the "DMD strength" to 70% to make it more useful with the config. I have two other versions, rank 64 and rank 128, but I haven't tested them. This version should be pretty complete, but I'll upload the lighter ones tomorrow. See you tomorrow!

Files with SHA256: 077cea198a35df3c8f9b3b04bef2fd46a16c08b78aea4dc7bd5e83af41d6a984