<h1 id="the-fallen-angel-anime-character-generation-via-rise-scheduler-1zsk0sq61">🧪 The Fallen Angel — Anime Character Generation via RISE Scheduler</h1><h2 id="abstract-lqzziogba">Abstract</h2>The Fallen Angel is a custom-trained anime checkpoint designed for simplified prompt input while maintaining high-fidelity, lovable anime girl characters. The model introduces a novel training scheduler—RISE (Rise Inverse Stable Evolution)—as part of ongoing research into scheduler behavior within DreamBooth. This is the 89th training iteration in my series, and the resulting model performs well with simple poses, high-resolution upscaling, and minimal prompt engineering.<h2 id="introduction-6wnghkqdr">Introduction</h2>Anime generation in diffusion models often requires complex prompt crafting and intensive tuning. The Fallen Angel seeks to simplify this process by introducing a bias towards well-composed, single-character (1girl) outputs with minimal prompts. By emphasizing simplicity and character fidelity, this model is particularly suited for enthusiasts and creators who prefer consistent, charming results without heavy prompt engineering.<h2 id="methodology-kekbjy6ur">Methodology</h2><h3 id="model-focus:-8klp5ghlm">Model Focus:</h3><ul><li>Style: Anime (1girl, clean pose, expressive characters)</li><li>Prompt Efficiency: Reduced prompt complexity</li><li>Orientation: Primarily 512x768 (portrait)</li></ul><h3 id="training-details:-2o1tt5o8s">Training Details:</h3><ul><li>Training Iteration: #89</li><li>Custom Scheduler: RISE (Rise Inverse Stable Evolution) <a target="_blank" rel="ugc" href="https://www.hitoha.moe/rise-algorithm-for-dreambooth/">Read more on the RISE algorithm</a></li><li>Optimized for portrait compositions and emotional expressiveness</li></ul><h2 id="inference-configuration-miyo0a6wo">Inference Configuration</h2>Recommended Settings:<ul><li>Sampling Method: DDIM or DDIM CFG++</li><li>Sampling Steps: 30</li><li>Resolution: 512x768</li><li>CFG Scale: 12</li></ul>Prompt Prefix:<pre><code>masterpiece, high quality, 1girl, &lt;name&gt;, &lt;dress&gt;, &lt;facial expression&gt;, ...</code></pre>Negative Prompt:<pre><code>worst quality, low quality, blurry, monochrome, sketch, lineart, bad anatomy, deformed, extra limbs, missing limbs, poorly drawn face, bad proportions, ugly, disfigured, bad lighting</code></pre><h2 id="upscaling-(hires.-fix)-hc0muuhh5">Upscaling (Hires. Fix)</h2>LDSR Upscaler (Recommended):<ul><li>Denoising Strength: 0.35 - 0.75</li><li>Upscale Factor: 1.5x</li></ul>LDSR has shown the best results with this model, especially for preserving fine line art and facial features.<h2 id="results-vffchgot3">Results</h2>The model performs exceptionally well in generating charming anime girls with appealing facial expressions and balanced compositions. It demonstrates high prompt responsiveness and low failure rate even at low sampling steps.<h2 id="conclusion-2j969sv3q">Conclusion</h2>The Fallen Angel demonstrates that simplicity in generation can coexist with quality, thanks in part to the newly proposed RISE scheduler. It’s ideal for those who want efficient, high-quality anime generations without prompt fatigue.

Fallen Angel

I added 3 new anime of this seasons, source is from streaming site up to 2nd Episode, screenshot video frame and properly upscale and remove H.264 artifacts. (Ripping from frames Blu-ray is recommended for up-coming SDXL)Suki na Ko ga Megane wo Wasureta<ol><li>Mie Ai</li></ol>Uchi no Kaisha no Chiisai Senpai no Hanashi<ol><li>Katase Shiori</li></ol>Yumemiru Danshi wa Genjitsushugisha<ol><li>Aizawa Rena</li><li>Ashida Kei</li><li>Inatomi Yuyu</li><li>Natsukawa Aika</li><li>Sajou Kaede</li><li>Shinomiya Rin</li></ol>School Uniform for Yumemiru Danshi wa Genjitsushugisha prompt as follow:<ol><li>kouetsu uniform (Global, random cardigan and tie/bow)</li><li>grey cardigan (aizawa rena)</li><li>black cardigan (sajou kaede)</li><li>cardigan (natsukawa aika)</li><li>red tie (natsukawa aika &amp; aizawa rena)</li><li>green tie (sajou kaede &amp; shinomiya rin)</li><li>red bow (ashida kei)</li><li>blue bow (inatomi yuyu)</li></ol>Special prompt for Mie Ai (Suki na Ko ga Megane wo Wasureta)<ol><li>glasses (Mie Ai will wearing glasses)</li><li>squinting (Mie Ai will squinting the eyes)</li><li>glasses, squinting (Mie Ai will do both, usually her in deep expression)</li></ol>Expect this checkpoint will be get better once ripping Up to 12th Episode, ripping video frame is common practice most anime checkpoint here.

Files with SHA256: 40ed9edc7a751224ed42d74875169ef86a18d3b7a61cd0879925a6d6043b3020