<h1 id="noctiluxe-usage-guide-love">Noctiluxe – Usage Guide ♥️</h1><h2 id="overview">Overview</h2><ul><li>Noctiluxe is a custom-merged CKPT model primarily optimized for anime, stylized, and ILL-inspired artwork. It is not strongly trained for realistic imagery, so generating consistent realism can be difficult. For anime and expressive stylized output, it performs exceptionally well.</li></ul><h3 id="recommended-workflow-in-comfyui">Recommended Workflow in ComfyUI</h3>Preview:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/3538e606-be57-4cfe-a942-c4c4cd19d87e/width=525/3538e606-be57-4cfe-a942-c4c4cd19d87e.jpeg" />Drag this <a target="_blank" rel="ugc" href="https://iili.io/fus3Adu.png">workflow image</a> into your ComfyUi[Update 12-14-25]<h3 id="1.-initial-generation-phase">1. Initial Generation Phase</h3><ul><li>Sampler: euler_ancestral</li><li>Scheduler: normal</li><li>Steps: 20</li><li>CFG: 5.0 - 7.0</li></ul><h3 id="2.-hi-res-fix-via-image-to-image">2. Hi-res fix via Image-to-Image</h3><ul><li>Scale: 1.5 - 2</li><li>Method: nearest-exact</li><li>Sampler: euler_ancestral</li><li>Scheduler: normal</li><li>Denoise Strength: 0.30</li></ul>This refines details while preserving the original composition and style.<h3 id="tags-and-style-triggers">Tags &amp; Style Triggers</h3>Noctiluxe responds well to tags from:<a rel="ugc" href="https://danbooru.donmai.us/">Danbooru</a>positive:<pre><code>lazypos, masterpiece, best quality, detailed,</code></pre>negative:<pre><code>worst aesthetic, worst quality, lowres, scan artifacts, signature</code></pre>Anime aesthetics are its strongest point.Realistic styles may require experimentation and are not guaranteed.About the MergeThe model was merged from various other models and LoRAs based on personal preference.Exact sources and merge values are not documented.ContactIf you have questions or need help, feel free to reach out.

Noctiluxe

<ul><li>Improved reliability – the model now runs more stable and consistent than before.</li><li>Significantly enhanced renderings – generated outputs are clearer, more detailed, and overall higher quality.</li><li>EULA_A compatibility – creating with eula_a now works reliably without the previous hiccups.</li></ul>Personal note: I’m honestly surprised at how good this merge turned out. The improvements are not only noticeable but make working with the model much smoother and more enjoyable.

noctiluxe_v2.safetensors

creator: NovamindNoctiluxe – Usage Guide ♥️OverviewNoctiluxe is a custom-merged CKPT model primarily optimized for anime, stylized, and ILL-inspired artwork. It is not strongly trained for realistic imagery, so generating consistent realism can be difficult. For anime and expressive stylized output, it performs exceptionally well.Recommended Workflow in ComfyUIPreview:<a target="_blank" rel="noopener noreferrer nofollow" href="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/3538e606-be57-4cfe-a942-c4c4cd19d87e/3538e606-be57-4cfe-a942-c4c4cd19d87e.jpeg">https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/3538e606-be57-4cfe-a942-c4c4cd19d87e/3538e606-be57-4cfe-a942-c4c4cd19d87e.jpeg</a>Drag this <a target="_blank" rel="ugc" href="https://iili.io/fus3Adu.png">workflow image</a> into your ComfyUi1. Initial Generation Phase<ul><li>Sampler: DDPM</li><li>CFG Rescale: 0.4</li><li>Steps: 20</li><li>CFG: 6.5</li></ul>DDPM tends to produce the most pleasing and stable base drafts for this model.2. Upscaling the First Draft<ul><li>Scale: 1.5×</li><li>Method: Bilinear</li></ul>This step creates a cleaner and more detailed canvas before refining the image.3. Refining via Image-to-Image<ul><li>Sampler: Euler_ancestral</li><li>Denoise Strength: 0.40</li></ul>This refines details while preserving the original composition and style.Tags &amp; Style TriggersNoctiluxe responds well to tags from:<ul><li>SDXL</li><li>Pony</li><li>ILL</li></ul>pos:lazypos, masterpiece, best quality, detailed,neg:worst aesthetic, worst quality, lowres, scan artifacts, signatureAnime aesthetics are its strongest point.Realistic styles may require experimentation and are not guaranteed.About the MergeThe model was merged from various other models and LoRAs based on personal preference.Exact sources and merge values are not documented.ContactIf you have questions or need help, feel free to reach out.

Files with SHA256: 565ea38394b550e43a0ac61014513056179fa80774491e1546d9b4a02dcbd949