This is the initial release of the LoRA trained by @1041uuu (TOYOI Yuuta) pixel artworks.⚠️ Pixel perfection is NOT guaranteed. This LoRA is incompatible with the <a target="_blank" rel="ugc" href="https://civarchive.com/models/2299499">Pixelate x4 VAE</a>. For best results, I recommend generating at 512x512 or higher using base-v01 with 30 steps / 5 CFG and <s>Euler</s>ER_SDE / Simple. However, please note that it still will not achieve flawless pixel perfection.I attempted to clean up the messy or broken pixels using post-processing tools like <a target="_blank" rel="ugc" href="https://github.com/KohakuBlueleaf/PixelOE#readme">PixelOE</a> and <a target="_blank" rel="ugc" href="https://github.com/HSDHCdev/ComfyUI-AI-Pixel-Art-Enhancer#readme">ComfyUI-AI-Pixel-Art-Enhancer</a>, but in my testing, it was impossible to fix them without losing the original aesthetic and atmosphere of the style. Therefore, for this v0.0.0 release, I have decided to publish it as-is, in its "pixel-imperfect" state.<h3 id="recommended-settings">Recommended Settings</h3><ul><li>Recommended Weight: <code>1.0</code></li><li>Optional Positive Prompts: <code>jaggy lines, dithering, no lineart, limited palette</code></li><li>Optional Negative Prompts: <code>jpeg artifacts, isometric, letterboxed, pillarboxed, dated</code></li></ul>

@1041uuu TOYOI Yuuta Style

Update Notes: v0.1.0<ul><li>Dataset Changes: Increased the training image upscale factor from x2 to x8.</li><li>Training Adjustments: Tweaked the configurations to be more prone to overfitting.<ul><li>Settings: <code>AdamW8bit</code>, <code>unet_lr=1e-4</code>, <code>dim=32</code>, <code>alpha=32</code>, <code>LR Scheduler: cosine_with_restarts (100 steps, cycles=1)</code>, <code>gradient_accumulation_steps=1</code>.</li></ul></li></ul>What to expect in v0.1.0:<ul><li>Unlike v0.0.0 and v0.0.1, generating at lower resolutions will now result in very low-pixel-count pixel art (giving a true micro-pixel aesthetic).</li><li>Because of the high-overfitting training settings, compatibility with other LoRAs has decreased, and you may encounter more style conflicts.</li><li>On the bright side, the generation results at resolutions of 1024 or higher have improved drastically—the difference is night and day!</li></ul>

Files with SHA256: bdb0c0ea933f1ea3d2264eeb954b28d9c2c29330caec1c09dfceb4ff6fa13f40