0xSoul 发表于 2023-6-7 19:24:49

[SD 1.5][LORA]Figma Anime Figures figma

<p><strong>Note</strong>: I did not train these LoRA models myself. Original source: <a target="_blank" rel="ugc" href="https://huggingface.co/stma/lora-dump">https://huggingface.co/stma/lora-dump</a></p><h2><strong>Recommended settings</strong></h2><p><strong>Quick start</strong></p><ul><li><p>Model: <a target="_blank" rel="ugc" href="https://civitai.com/models/4437/abyssorangemix2-sfw">AbyssOrangeMix2</a></p></li><li><p>LoRA weight: 0.6 (UNet: 0.6, TEnc: 0.25)</p></li><li><p><code>, , (3dcg), ((octane render)), </code> in positive prompt for realism</p></li><li><p><code>simple background, grey background</code> in negative prompt</p></li><li><p>CLIP skip: 2</p></li><li><p>Sampler: DPM++ SDE Karras (Euler a recommended for other models)</p></li><li><p><a target="_blank" rel="ugc" href="https://huggingface.co/datasets/gsdf/EasyNegative">EasyNegative </a>in negative prompt</p></li></ul><p><strong>Usable</strong></p><ul><li><p>LoRA weight range: 0.2 - 1.15 (UNet: 0.35 - 1.15 TEnc: 0.2 - 1.0)</p></li></ul><p>UNet is required for the style. UNet over 0.65 will look like an anime figure product. Too high UNet weight will melt details.</p><p>Lower TEnc weight will improve generalization. High TEnc will affect composition to look more like an anime figure product photo. Too high will fry the details.</p><p>Start with lower weight and increase as necessary. Requires more weight when mixed with TI embeddings and other LoRAs.</p><h2><strong>Figma</strong></h2><p>Trained on 5,746 images scraped from GoodSmile's international catalogue, including the discontinued products. All images were tagged with <code>figma</code> first and the product name with <code>figma</code> stripped out second, so an image from a product named "figma Hatsune Miku" would be tagged <code>figma, hatsune miku</code>. All other tags came from the WD1.4 tagger. <code>3d</code> and <code>realistic</code> tags seem to work well with it.</p><p>The alpha16 version is better at generalizing, while the 7e-5 version is better at capturing the figure-with-transparent-stand look.</p><h2><strong>Example images</strong></h2><h3>Model</h3><ul><li><p><a target="_blank" rel="ugc" href="https://huggingface.co/gsdf/Counterfeit-V2.5">Counterfeit-V2.5 </a>(TODO: update example images to <a target="_blank" rel="ugc" href="https://civitai.com/models/4437/abyssorangemix2-sfw">AbyssOrangeMix2</a>)</p></li></ul><h3>Textual inversion</h3><ul><li><p><a target="_blank" rel="ugc" href="https://civitai.com/models/4972/roxy-migurdia-ti">Roxy Migurdia</a> (character) <code>roxy-migurdia</code></p></li><li><p><a target="_blank" rel="ugc" href="https://civitai.com/models/5537/megumin-ti">Megumin</a> (character) <code>megumin_1000</code></p></li><li><p><a target="_blank" rel="ugc" href="https://civitai.com/models/3036/charturner-character-turnaround-helper">Char Turner </a>(composition) <code>chartuner</code></p></li><li><p><a target="_blank" rel="ugc" href="https://huggingface.co/datasets/gsdf/EasyNegative">EasyNegative </a><code>EasyNegative</code></p></li><li><p><a target="_blank" rel="ugc" href="https://huggingface.co/datasets/Nerfgun3/bad_prompt">Bad Prompt </a><code>bad_prompt</code></p></li></ul><h3>Additional LoRA</h3><ul><li><p><a target="_blank" rel="ugc" href="https://civitai.com/models/4831/yelan-lora-collection-of-trauters">Yelan</a> (charater) <code>yelan \(genshin impact\)</code></p></li><li><p><a target="_blank" rel="ugc" href="https://civitai.com/models/5126/torino-aqua-style-lora">Torino Aqua</a> (style) <code>torino aqua</code></p></li></ul>
页: [1]
查看完整版本: [SD 1.5][LORA]Figma Anime Figures figma