0xSoul 发表于 2023-6-7 22:23:04

[Other][Hypernetwork]SXZ Cinematic 800 Strong Enough

<p>This hypernetwork was trained on Anything V3 so it should work with almost any anime-based model.</p><p>Every sample image was made with AbyssOrange_hard.</p><p><strong>I recommend 0.6 strenght for most generations.</strong></p><p>There is a few versions of hypernetwork, and each of it have different shading and affect on whole image with different strength.</p><img src="https://imagecache.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/f4077226-7123-445d-82ca-a231e8e51c00/width=525" /><p><br />Base dataset was 4K screenshots from Batman v Superman cropped to 512, so with more strenght of hypernetwork images will colorized more like color palette of this movie, and have more film grain with bokeh.<br /></p><img src="http://127.0.0.1:7860/file=E:/STABLE%20DIFFUSION/outputs/txt2img-grids/xy_grid-0041-2612120270-upper%20body%20portrait,%201girl,%20short%20black%20hair,%20red%20atmosphere,%20fantasy,%20dark%20aesthetic,%20radiant%20colors,%20tall,%20medium%20breasts,%20mas.png" /><p><br /><strong>Sometimes it may conflict with LORA's styles</strong> so you may need to decrease strength of hypernetwork or LORA, also order of use matters in new Automatic 1111 interface, so maybe you will ned to experiment with stuff a lil bit.</p><img src="https://imagecache.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/eafc1ab6-d4ce-45f6-e10f-b13c054d0500/width=525" /><p><strong>Also if you have any info how to train hypernetworks, LORA's and TI properly, please share your knowledge.</strong></p>
页: [1]
查看完整版本: [Other][Hypernetwork]SXZ Cinematic 800 Strong Enough