<h1 id="heading-7264">Kohaku XL beta</h1><h3 id="heading-7265">An anime SDXL model trained on 1.5M images.</h3> Notice: Further experiments have showed that beta7 is generally better than beta7pro. Beta7pro have some mosaic artifact (more significant than beta7) and beta7 have better txt-img alignment. So I rearrange the order of models. <hr /><h2 id="heading-7266">Introduction</h2>This model is resumed from [Kohaku-XL alpha](<a target="_blank" rel="ugc" href="https://civarchive.com/models/136389/kohaku-xl-alpha">Kohaku-XL alpha - nyan | Stable Diffusion Checkpoint | Civitai</a>) with 1.5M images and then merged with other models.<hr /><h2 id="heading-7267">Usage Details</h2>This model is very flexible on resolution, you can use the resolution you used in sd1.x/2.x to get normal result(like 512x768), you can also use the resolution that is more native for sdxl(like 896*1280) or even bigger (1024x1536 also ok for t2i).recommended negative prompt for anime style:<ul><li><code>photorealistic, 3d model, bad, worse, worst, ugly, bad anatomy, blurry, close-up, disembodied limb</code></li><li><code>photorealistic, text, icon, artist name, signature, twitter username, naked, nude, monochrome, blurry, bad anatomy, watercolor, oil painting</code></li><li><code>watercolor, oil painting, photo, deformed, realism, disfigured, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry</code></li></ul><hr /><h2 id="heading-957">Training Details</h2>Kohaku-XL beta5This model is trained on new-danbooru (danbooru images with id from 5,000,000~6,600,000) which have 1.48M images. This model is resumed from kohaku-xl alpha7 and then merged with NekoRayXL.Kohaku-XL base4 (haven't publish yet)This model is trained on new-danbooru (danbooru images with id from 5,000,000~6,600,000) which have 1.48M images. Thie model is resumed from sdxl-0.9 (due to some bad property in sdxl-1.0 which will affect finetuning). In the plan this model will be trained with 2epoch (about 94.5k steps).I haven't published this pretrained model yet.Kohaku-XL beta7This model is merged with base4 and beta 5, the formula is:<pre><code>beta(5+n) = beta5 + (n/4) * (base4 - sdxl0.9)</code></pre>So beta7 is <code>beta5 + 0.5 * (base4 - sdxl0.9)</code> Kohaku-XL beta7.1(7pro)as same as beta7 but use the finished base4 and 0.25 weight. Note: the base4 here is the 50k step version!! <h2 id="heading-958">Future Plan</h2>I will run training on base4(after it finish) in Mynefactory dataset or CyberMeow(alea31415)/Narugo1992's reg dataset.<hr /><h2 id="heading-7268">Acknowledgements</h2>Models<ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/136719/nekorayxl">NekoRayXL</a></li></ul>

Kohaku-XL beta

Use finished base4 to merge with beta5 (beta7.1 = beta5 + 0.25 * (base4 - sdxl))

kohakuXLBeta_beta69.safetensors

beta6.9

beta7

beta5

Kohaku XL betaAn anime SDXL model trained on 1.5M images.IntroductionThis model is resumed from [Kohaku-XL alpha](<a target="_blank" rel="ugc" href="https://civitai.com/models/136389/kohaku-xl-alpha">Kohaku-XL alpha - nyan | Stable Diffusion Checkpoint | Civitai</a>) with 1.5M images and then merged with other models.Usage DetailsThis model is very flexible on resolution, you can use the resolution you used in sd1.x/2.x to get normal result(like 512x768), you can also use the resolution that is more native for sdxl(like 896*1280) or even bigger (1024x1536 also ok for t2i).recommended negative prompt for anime style:<ul><li>photorealistic, 3d model, bad, worse, worst, ugly, bad anatomy, blurry, close-up, disembodied limb</li><li>photorealistic, text, icon, artist name, signature, twitter username, naked, nude, monochrome, blurry, bad anatomy, watercolor, oil painting</li><li>watercolor, oil painting, photo, deformed, realism, disfigured, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry</li></ul>Training DetailsKohaku-XL beta5This model is trained on new-danbooru (danbooru images with id from 5,000,000~6,600,000) which have 1.48M images. This model is resumed from kohaku-xl alpha7 and then merged with NekoRayXL.Kohaku-XL base4 (haven't publish yet)This model is trained on new-danbooru (danbooru images with id from 5,000,000~6,600,000) which have 1.48M images. Thie model is resumed from sdxl-0.9 (due to some bad property in sdxl-1.0 which will affect finetuning). In the plan this model will be trained with 2epoch (about 94.5k steps).I haven't published this pretrained model yet.Kohaku-XL beta7This model is merged with base4 and beta 5, the formula is:beta(5+n) = beta5 + (n/4) * (base4 - sdxl0.9)So beta7 is beta5 + 0.5 * (base4 - sdxl0.9)Kohaku-XL beta7.1as same as beta7 but use the finished base4 and 0.25 weight.Note: the base4 here is the 50k step version!!Future PlanI will run training on base4(after it finish) in Mynefactory dataset or CyberMeow(alea31415)/Narugo1992's reg dataset.AcknowledgementsModels<ul><li><a target="_blank" rel="ugc" href="https://civitai.com/models/136719/nekorayxl">NekoRayXL</a></li></ul>

[beta7-pro] Use finished base4 to merge with beta5
(beta7.1 = beta5 + 0.25 * (base4 - sdxl))

Files with SHA256: 51a0c178b7ad47966bdf3397e51a385edf151ed387ba31bab447a12071507933