This is an experimental checkpoint that combines some of the best realistic Illustrious models with bigASP 2.5. It worked.BTW, <a target="_blank" rel="ugc" href="https://civarchive.com/models/2068592/snakebite-2">v2.4 is now available here!</a> Give it a try! ❤️ If you like Snakebite, you can help offset the cost of training:<a target="_blank" rel="ugc" href="https://ko-fi.com/liftweights">Buy liftweights a Coffee</a> <h2 id="why-it-matters">Why it matters</h2> bigASP has great prompt adherence but it is wildly inconsistent as far as style and composition. It feels like a base model with untapped potential. Dialing in the right settings is like trying to solve a Rubik's Cube. Illustrious models, on the other hand, lose many concepts when going 3D. They're hit-or-miss even with fairly popular booru tags. But the lighting and composition of these weights are still 👌 I wanted to see if we can get the best of both worlds, and it turns out we kinda can! Careful block merging is the key. We can inject bigASP's output_blocks.0 to acquire much of its conceptual knowledge. Adding its middle_block.2 seems to reduce anatomical issues (otherwise, you'll get a lot of extra arms and fingers.)<h2 id="first-impressions">First impressions</h2><ul><li>Best prompt adherence for a realistic SDXL model I've ever seen. And no, not just for smut. (But especially for smut)</li><li>Compatible with booru tags as well as natural language. I think a mixed prompt approach is best: I use about 70% tags (no underscores) and 30% natural language in a prompt.</li><li>Understands bigASP's style tags to an extent, including stuff like masterpiece quality and 35mm.</li><li>It *feels* like something new and worth exploring. Don't sleep on bigASP!</li></ul><h2 id="drawbacks">Drawbacks</h2><ul><li>Illustrious 2.0 models support resolutions up to 1024x1440 or 1024x1536 without horrific stretching of anatomy (e.g. longcat-type torsos), but bigASP's optimal resolution is only 832x1216... and I don't recommend going above that in Snakebite. If you do, the anatomy will be mostly okay (which is surprising), but image composition becomes very odd and unpleasant.</li><li>Since we're in a strange new latent space, your existing LoRAs won't work very well. But they're worth retraining.</li></ul><hr /><h2 id="recommended-settings">Recommended Settings</h2>The Turbo variant is better for inference. It's super fast and has slightly improved aesthetics. The non-turbo version is useful for finetuning, and it can produce nice textures if you don't mind waiting 25+ steps. <h3 id="turbo">Turbo</h3><ul><li>8 or 9 steps</li><li>LCM sampler</li><li>CFG 1</li><li>Custom sigmas below, or simple<code> </code></li></ul><h3 id="full">Full</h3><ul><li>20 to 28 steps</li><li>Euler ancestral sampler</li><li>CFG 3 to 4</li><li>Custom sigmas below, or simple </li></ul>Custom sigma curve (you can use comfyui-kjnodes to apply): <pre><code>15, 8, 4, 2, 2, 1, 0.4, 0.2, 0</code></pre> If you're getting mangled limbs, you can often salvage the image by adjusting the first few values of your sigma curve. Here's one that is more stable for certain prompts:<pre><code>14, 5, 2, 2, 2, 1, 0.4, 0.2, 0</code></pre><hr />If you're still getting body horror, you can try the following quality tags (you'll need <a target="_blank" rel="ugc" href="https://github.com/pamparamm/ComfyUI-ppm">ComfyUI-ppm</a> to apply negative weights to your positive prompt): <pre><code>masterpiece quality, realistic photo, (worst quality,:-1) (mutated,:-1)</code></pre>Snakebite is very responsive to stylistic terms, especially by IL standards. Keep the extra "fluff" to a minimum - almost every token I've tried has a significant impact on the picture.Finally, I suggest trying the CLIPAttentionMultiply node. If you boost the <code>q</code> and <code>v</code> parameters, it will effectively cause your image to become more "Illustrious-like": cleaner, more stable, but less realistic and (usually) less adherent to the prompt. Set both values to 3 for a very clean image.BIGASP'S CLIP IS NOW IN PLAY!In versions 1.3 and up, Snakebite includes a little of bigASP's CLIP, which means you can take advantage of more style prompts. Experiment with different terms to see what works. Personally, I keep it simple - this will usually improve your image without any side effects:<pre><code>high quality, sharp focus</code></pre><hr /><h2 id="which-version-is-for-me">Which version is for me?</h2>If you're wondering which version of the model to use, here's a TL;DR:<ul><li>v1.4 = next-level realism, jaw-dropping textures, very stable, slightly less vibrant than previous versions and less capable of non-photographic images</li><li>v1.3 = good anatomy, good backgrounds, good coherence</li><li>v1.2 = best punchy colors</li><li>v1.1 = most influence from bigASP (excluding CLIP), dull colors, a failed experiment TBH</li><li>v1.0 = impressively creative but very unstable</li></ul><hr />If you like the model or use it for further finetuning, please let me know! I'd love to see the results. 💪

Snakebite

Big update here! Excited to share it with you.I was satisfied with the overall balance between IL and bigASP, so I took my first crack at aesthetic tuning for v1.4. It's capable of pretty amazing realism now. I've managed to retain nearly all of Snakebite's knowledge from IL while unearthing some incredible textures that were hidden in bigASP's latent space. Usually, you have to choose between textures and prompt adherence. I think we can finally have our cake and eat it too.I also implemented a few LoRAs that should assist with overall image composition and reduce unwanted limbs. Finally, I tweaked the balance of the CLIP just a smidge (it's almost entirely IL, with just the smallest hint of bigASP clip to help resolve anatomy.)In a nutshell: faces are more diverse, human anatomy is more stable, and the model's baseline aesthetic is more cinematic/realistic/texture-rich.Is it perfect? Nah, v1.4 lost a little of the punchy vibrance from previous versions despite my best efforts. It's a double-edged thing. The vibrance makes it possible to generate super-clean images that are otherwise beyond SDXL's reach, but it also makes things look a bit "CGI." I've tried keeping as much vibrance as possible while significantly boosting the realism. Having said that, you may prefer the aesthetic of v1.3 for illustrations and digital art.

Files with SHA256: 89d2da139ddb00ed40a190c584ad46b38219680ef956b2ed1d910b8d12b0517f