<ul><li>Negative Prompt:</li></ul><pre><code>notice lines, spoken heart,bad quality, worst quality, worst detail, sketch, censor, </code></pre><h1 id="dsharp-k:">dsharp k:</h1><a target="_blank" rel="ugc" href="https://twitter.com/Dsharp_K/">https://twitter.com/Dsharp_K/</a><a target="_blank" rel="ugc" href="https://bsky.app/profile/kddsyp.bsky.social">https://bsky.app/profile/kddsyp.bsky.social</a><a target="_blank" rel="ugc" href="https://www.patreon.com/Dsharp_k">https://www.patreon.com/Dsharp_k</a><a target="_blank" rel="ugc" href="https://gumroad.com/dsharp91">https://gumroad.com/dsharp91</a>New version updates:<ul><li>Greatly expanded the training dataset</li><li>Manually tagged specific images for better accuracy</li><li>Increased resolution parameters during training</li></ul>Current known issues:<ul><li>Poor performance with multiple characters</li><li>Inconsistent facial generation</li><li>Unwanted elements often appear (e.g. notice lines, spoken hearts)</li></ul>A fix is expected in the near future.<hr /><ul><li>Prompt Prefix:</li></ul><pre><code>masterpiece, best quality, amazing quality, </code></pre>Optional Prompt :Manual tagging has been performed on the following content.<pre><code>(((cum on male:1.4, cum on body:1.4, cum on penis:1.4,cum on pectorals:1.4)))</code></pre><pre><code>(((ejaculation,motion blur,afterimage,smear frame)))</code></pre><ul><li>Optional Negative Prompt:</li></ul><pre><code>monochrome, greyscale, signature, url, artist name, username, watermark, text, English text, </code></pre>

Dsharp k Style [ILXL]

{ "engine": "kohya", "unetLR": 1, "clipSkip": 2, "loraType": "lora", "keepTokens": 1, "networkDim": 32, "numRepeats": 4, "resolution": 1024, "lrScheduler": "cosine", "minSnrGamma": 5, "noiseOffset": 0.1, "targetSteps": 1908, "enableBucket": true, "networkAlpha": 16, "optimizerType": "Prodigy", "textEncoderLR": 1, "maxTrainEpochs": 27, "shuffleCaption": true, "trainBatchSize": 6, "flipAugmentation": true, "lrSchedulerNumCycles": 3}

dsharp_k_style.safetensors

ILXL0.1

ILXL0.1_2.1

ILXL0.1_2

<h1 id="dsharp-k:-3r8y680m2">dsharp k:</h1><a target="_blank" rel="ugc" href="https://twitter.com/Dsharp_K/">https://twitter.com/Dsharp_K/</a><a target="_blank" rel="ugc" href="https://bsky.app/profile/kddsyp.bsky.social">https://bsky.app/profile/kddsyp.bsky.social</a><a target="_blank" rel="ugc" href="https://www.patreon.com/Dsharp_k">https://www.patreon.com/Dsharp_k</a><a target="_blank" rel="ugc" href="https://gumroad.com/dsharp91">https://gumroad.com/dsharp91</a>Optional Prompt :<pre><code>2024, 2023, </code></pre>Negative Prompt :<pre><code>url, artistname, username, text, English text, bad quality,worst quality,worst detail,sketch,censor,</code></pre>This is a simple attempt. The training data and tags are all from e621, so they are not really excellent training pictures. I haven't changed its tag either. This model has not been completely tested, but it looks good so far. If this effect is good enough, maybe I don't need to look for better training pictures.After testing, I think this model is not well trained and will be updated in the future.

Dsharp_k Style [Illustrious]

&lt;p&gt;{&lt;/p&gt;&lt;p&gt;  &#34;engine&#34;: &#34;kohya&#34;,&lt;/p&gt;&lt;p&gt;  &#34;unetLR&#34;: 1,&lt;/p&gt;&lt;p&gt;  &#34;clipSkip&#34;: 2,&lt;/p&gt;&lt;p&gt;  &#34;loraType&#34;: &#34;lora&#34;,&lt;/p&gt;&lt;p&gt;  &#34;keepTokens&#34;: 1,&lt;/p&gt;&lt;p&gt;  &#34;networkDim&#34;: 32,&lt;/p&gt;&lt;p&gt;  &#34;numRepeats&#34;: 4,&lt;/p&gt;&lt;p&gt;  &#34;resolution&#34;: 1024,&lt;/p&gt;&lt;p&gt;  &#34;lrScheduler&#34;: &#34;cosine&#34;,&lt;/p&gt;&lt;p&gt;  &#34;minSnrGamma&#34;: 5,&lt;/p&gt;&lt;p&gt;  &#34;noiseOffset&#34;: 0.1,&lt;/p&gt;&lt;p&gt;  &#34;targetSteps&#34;: 1908,&lt;/p&gt;&lt;p&gt;  &#34;enableBucket&#34;: true,&lt;/p&gt;&lt;p&gt;  &#34;networkAlpha&#34;: 16,&lt;/p&gt;&lt;p&gt;  &#34;optimizerType&#34;: &#34;Prodigy&#34;,&lt;/p&gt;&lt;p&gt;  &#34;textEncoderLR&#34;: 1,&lt;/p&gt;&lt;p&gt;  &#34;maxTrainEpochs&#34;: 27,&lt;/p&gt;&lt;p&gt;  &#34;shuffleCaption&#34;: true,&lt;/p&gt;&lt;p&gt;  &#34;trainBatchSize&#34;: 6,&lt;/p&gt;&lt;p&gt;  &#34;flipAugmentation&#34;: true,&lt;/p&gt;&lt;p&gt;  &#34;lrSchedulerNumCycles&#34;: 3&lt;/p&gt;&lt;p&gt;}&lt;/p&gt;

Files with SHA256: 175b472d7cf6820fca8165fe1abc74e717553aa0a54b55f31d4cc60441ad26a7