<h1 id="wai-illustrious-sdxl-hswq-enhanced">WAI-illustrious-SDXL HSWQ Enhanced</h1><h3 id="fast-like-fp8.-stable-like-full-precision.">Fast Like FP8. Stable Like Full Precision.</h3><p>A performance-focused merge built on the HSWQ FP8 edition of WAI-illustrious-SDXL v17.0, enhanced with DMD2, Stabilizer IL, and SPO-SDXL_4k-p_10ep.</p><p>This model was created with a simple goal:</p><p><strong>Deliver high-quality anime image generation while maintaining the speed and efficiency advantages of FP8 quantization.</strong></p><p>By combining modern optimization techniques with carefully selected merges, this checkpoint aims to provide strong prompt adherence, improved stability, and excellent visual quality even at relatively low step counts.</p><hr /><h1 id="what's-inside">What's Inside</h1><h3 id="base-model">Base Model</h3><ul><li><p>WAI-illustrious-SDXL v17.0</p></li></ul><h3 id="additional-merges">Additional Merges</h3><ul><li><p>DMD2</p></li><li><p>Stabilizer IL</p></li><li><p>SPO-SDXL_4k-p_10ep</p></li></ul><p>Each component was selected to improve a different aspect of generation quality:</p><ul><li><p>DMD2 helps achieve stronger results at lower sampling steps.</p></li><li><p>Stabilizer IL improves generation consistency and reliability.</p></li><li><p>SPO-SDXL_4k-p_10ep enhances composition, structure, and detail retention.</p></li></ul><p>The result is a model optimized for both rapid iteration and everyday image generation.</p><hr /><h1 id="what-is-hswq">What is HSWQ?</h1><p>HSWQ (Hybrid Sensitivity-Weighted Quantization) is an advanced quantization technique designed to reduce model size and VRAM requirements while preserving image quality.</p><p>Traditional FP8 quantization often applies the same level of compression throughout the entire network. While efficient, this can sometimes introduce quality degradation in layers that are particularly important for image generation.</p><p>HSWQ takes a smarter approach.</p><p>Instead of treating every layer equally, it analyzes layer sensitivity and applies quantization more selectively. Critical layers receive greater protection while less sensitive components can be compressed more aggressively.</p><p>This allows the model to retain much of the visual quality and prompt understanding of the original checkpoint while benefiting from the efficiency of FP8.</p><h3 id="benefits-of-hswq">Benefits of HSWQ</h3><ul><li><p>Lower VRAM requirements</p></li><li><p>Faster loading times</p></li><li><p>Faster inference</p></li><li><p>Better quality retention than conventional FP8 approaches</p></li><li><p>More accessible generation on a wider range of hardware</p></li></ul><p>Think of HSWQ as a "smart FP8" approach that preserves precision where it matters most.</p><hr /><h1 id="key-features">Key Features</h1><h3 id="fast-generation">⚡ Fast Generation</h3><p>Designed around an FP8 workflow for efficient image generation and reduced memory consumption.</p><h3 id="improved-prompt-adherence">🎯 Improved Prompt Adherence</h3><p>Additional merges help maintain character details, clothing descriptions, and scene composition more accurately.</p><h3 id="enhanced-visual-quality">🖼 Enhanced Visual Quality</h3><p>Improved detail retention, stronger compositions, and more consistent image structure.</p><h3 id="low-step-performance">🚀 Low-Step Performance</h3><p>Produces attractive results even with relatively low step counts, making it ideal for online generators and rapid experimentation.</p><h3 id="balanced-workflow">💡 Balanced Workflow</h3><p>Rather than maximizing a single metric, this model focuses on balancing:</p><ul><li><p>Speed</p></li><li><p>Stability</p></li><li><p>Detail</p></li><li><p>Prompt responsiveness</p></li><li><p>Resource efficiency</p></li></ul><hr /><h1 id="recommended-settings">Recommended Settings</h1><h2 id="online-generation-fast-workflow">Online Generation / Fast Workflow</h2><ul><li><p>CFG Scale: 1.0</p></li><li><p>Steps: 10</p></li><li><p>Sampler: Euler a</p></li><li><p>Resolution: 1024×1024</p></li></ul><p>These settings were the primary target during testing and offer an excellent balance between speed and quality.</p><h2 id="higher-quality-generation">Higher Quality Generation</h2><ul><li><p>CFG Scale: 2.0–3.0</p></li><li><p>Steps: 15–20</p></li><li><p>Resolution: 1024×1024 or higher</p></li></ul><hr /><h1 id="recommended-use-cases">Recommended Use Cases</h1><p>This model performs particularly well for:</p><ul><li><p>Anime illustrations</p></li><li><p>Character portraits</p></li><li><p>Full-body character artwork</p></li><li><p>Fantasy scenes</p></li><li><p>Vibrant lighting</p></li><li><p>Expressive characters</p></li><li><p>Detailed costumes</p></li><li><p>Dynamic compositions</p></li></ul><hr /><h1 id="notes">Notes</h1><p>This is a merge model designed to provide a practical balance between performance and image quality.</p><p>If you want the absolute fastest workflow possible while maintaining attractive anime-style outputs, this model was built with that goal in mind.</p><p>Whether you're generating locally on limited hardware or using online generators with restricted resources, this checkpoint aims to make high-quality image creation more accessible.</p><hr /><h1 id="credits">Credits</h1><p>Base Model</p><ul><li><p>WAI-illustrious-SDXL v17.0</p></li></ul><p>Merged Components</p><ul><li><p>DMD2</p></li><li><p>Stabilizer IL</p></li><li><p>SPO-SDXL_4k-p_10ep</p></li></ul><p>Special thanks to all original model creators and researchers whose work made this merge possible.</p><p>Please respect the licenses and usage terms of all source models and components.</p>

Files with SHA256: 74baa0d6ee9690173d5f88c37f33ae7067f52020c7982eacbc765a6b6a5dc9bc