<h2 id="overview-dvn49alwb">Overview</h2>Introducing Rae Diffusion XL , an enhanced iteration based on the Animagine XL 3.1 model, specifically fine-tuned for generating stunning anime-style artwork. Rae Diffusion XL is meticulously optimized to excel in depicting anime characters, pushing the boundaries of creativity. <hr />Enjoying my work? Consider supporting the development of my models by buying me a coffee on <a target="_blank" rel="ugc" href="https://ko-fi.com/raelina">Ko-fi</a>Also check out my other models on Mage: <a target="_blank" rel="ugc" href="https://www.mage.space/play/55ee9a305fc7e3572509def13acbefbf">Raepix Focus XL</a> / <a target="_blank" rel="ugc" href="https://www.mage.space/play/08df2e9b11b4d9587dcbb6846a9dda11">Raepix Focus XL Lightning</a> / <a target="_blank" rel="ugc" href="https://www.mage.space/play/41658a734e3918e89e700c76f103ffc7">Raevitamix Realistic</a> / <a target="_blank" rel="ugc" href="https://www.mage.space/play/b1586a8de40e97411fdcda1fe03ceb60">Raevista XL</a><hr /><h2 id="anime-series-1fjh37139">Anime Series</h2>New additional anime series that have been trained:V1<ul><li>Fairy Tail</li><li>Gabriel Dropout</li><li>Blend S</li><li>Nisekoi</li><li>Youkoso Jitsuryoku Shijou no Kyoushitsu e</li><li>Kanojo Okarishimasu</li><li>Saenai Heroine no Sodatekata</li><li>Mirai Nikki</li><li>Guilty Crown</li></ul>Wildcard character available to download <a target="_blank" rel="ugc" href="https://huggingface.co/Raelina/Rae-Diffusion-XL/tree/main/wildcard">here</a>V2<ul><li>Zombie land saga</li><li>Yuru yuri</li><li>To love-ru</li><li>Amagi brilliant park</li><li>Avatar legends</li><li>Highschool of the dead</li><li>Infinite stratos</li><li>Inuyasha</li><li>Kara no kyoukai</li><li>Monster musume no iru nichijou</li><li>Seishun buta yarou</li><li>Snk</li><li>Tekken</li><li>Wuthering waves</li></ul>Wildcard character for V2 is available to download <a target="_blank" rel="ugc" href="https://huggingface.co/Raelina/Rae-Diffusion-XL-V2/tree/main/wildcard">here</a>*Note: Wildcards are automatically generated from metadata JSON. Some characters might not have been properly generated due to the limited dataset available. You may need more specific prompts, such as hair color, eye color, accessories, costume, etc.<h2 id="tag-ordering-dzvmqq5vh">Tag Ordering</h2>For optimal results, it's recommended to follow the structured prompt template because we train the model like this:<pre><code>1girl/1boy, character name, from which series, everything else in any order. </code></pre><h2 id="recommended-settings-nkyfi634z">Recommended settings</h2><ul><li>Positive prompts:</li></ul><pre><code>masterpiece, best quality, very aesthetic, absurdres,</code></pre><ul><li>Negative prompts:</li></ul><pre><code>(low quality, worst quality:1.2), very displeasing, ugly, poorly drawn, signature, watermark,</code></pre><ul><li>CFG: 7</li><li>Sampling steps: 25 to 35</li><li>Sampler: Euler a</li><li>Supported Resolution:</li></ul><pre><code>1024 x 1024, 1152 x 896, 896 x 1152, 1216 x 832, 832 x 1216, 1344 x 768, 768 x 1344, 1536 x 640, 640 x 1536</code></pre><h2 id="hires.fix-setting-zvuo6l1or">Hires.fix Setting</h2><ul><li>Upscaler: <a target="_blank" rel="ugc" href="https://nmkd.de/?esrgan">4x_NMKD-YandereNeoXL</a></li><li>Hires step: 10-15</li><li>Denoising: 0.1-0.3 or 0.55 for latent upscaler</li></ul><h2 id="training-config-szi9n42q2">Training config</h2><ul><li>Hardware: 1x A100 80GB</li><li>Batch size: 48</li><li>Gradient Accumulation: 1</li><li>Epochs: 10</li><li>Learning Rate: 3e-6</li><li>Optimizer: Adafactor</li><li>Optimizer Args: (Scale Parameter: False, Relative Step: False, Warmup Init: False)</li><li>Scheduler: Constant with warmup</li><li>Warmup steps: 0.05</li><li>Noise offset: 0.0357</li></ul><h2 id="license-sp7jo3tn6">License</h2>Rae Diffusion XL now uses the <a target="_blank" rel="ugc" href="https://freedevproject.org/faipl-1.0-sd/">Fair AI Public License 1.0-SD</a> inherited from Animagine XL 3.1, compatible with Stable Diffusion models. Key points:<ol><li>Modification Sharing: If you modify Rae Diffusion XL, you must share both your changes and the original license.</li><li>Source Code Accessibility: If your modified version is network-accessible, provide a way (like a download link) for others to get the source code. This applies to derived models too.</li><li>Distribution Terms: Any distribution must be under this license or another with similar rules.</li><li>Compliance: Non-compliance must be fixed within 30 days to avoid license termination, emphasizing transparency and adherence to open-source values.</li></ol>

Rae Diffusion XL

<ul><li>Increased dataset to 60k (4x more than v1)</li><li>Improved training used (now uses A100 80gb)</li><li>Update the scoring of quality tags</li><li>Improved stability</li></ul>

Files with SHA256: 98951313064352e5ed816868f7ae13bd9ad8acd932b5c2d28a87d1819357ef30