I hope you all enjoy the LoRA. Remember you can use my LoRAs in your checkpoint models long as you give credit. I cannot guarantee this works the same on every model.<hr />I'm working on a small AI Discord Community you can join here: <a target="_blank" rel="ugc" href="https://discord.gg/ehhMrD4PQT">https://discord.gg/ehhMrD4PQT</a><hr /><h1 id="disclaimer-what's-below-is-not-the-recommended-settings-you-will-have-to-figure-that-out-yourself!">DISCLAIMER - WHAT'S BELOW IS NOT THE RECOMMENDED SETTINGS, YOU WILL HAVE TO FIGURE THAT OUT YOURSELF!</h1><h2 id="(w.i.p)-v1.0-version-information:">(W.I.P) V1.0 Version Information:</h2><ul><li>Dataset includes images made from the previous dataset</li><li>New images from the V0.8 LoRA with a 5 LoRA, LoRA Mix</li><li>Some Images fixed with nano banana</li></ul><h3 id="for-those-who-it-may-concern.">For those who it may concern.</h3>The checkpoint used to generate images in the dataset: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2338192">Atlas Core V0.6</a> since it likes the LoRA somewhat<a target="_blank" rel="ugc" href="https://civarchive.com/models/1595884?modelVersionId=1805971"> </a>also used <a target="_blank" rel="ugc" href="http://civarchive.com/models/2179383?modelVersionId=1805971">Crystal VAE V3</a>.And these for the prompts and settings:Sampler EulerScheduler: NormalSteps: 20 / 25CFG: 4LoRAs Used:<ol><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2390241">Arctenox's Styles - Ouroboros | Series 2 [Illustrious]</a> on 67%</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2214514?modelVersionId=2695771">no100 (1号💙12/31東ノ-55a) style SDXL LoRA [Illustrious] | 3 Styles</a> on 27%</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2203490/arctenoxs-styles-sketch-fantasy-or-series-1-illustrious">Arctenox's Styles - Sketch Fantasy | Series 1 [Illustrious]</a> on 24%</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2128969?modelVersionId=2691971">VΞЯØX∄∆†H (Veroxh)</a> on 17%</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2406794?modelVersionId=2706125">Yrvne | UTdT69's Style</a> on 2%</li></ol>Positive:<pre><code>bsouororo, 5DFA1, yrvn, [PROMPT], masterpiece, amazing quality, best quality, highest quality, absurdres, highres, extremely detailed, very awa, official art, official style, highly detailed, anime_coloring, animification, [(artist:nt00:0.01), (artist:fuzichoco:0.01), (artist:swav:0.03), (artist:crote:0.03), (artist:unfairr:0.04), (fate_/(series/):0.2), (bleach:0.5), (black_clover:0.3)], v3rox_2th, IL_n0100 late style V2</code></pre>Negative:<pre><code>worst quality, low quality, normal quality, lowres, blurry, out of focus, jpeg artifacts, compression artifacts, ugly, bad anatomy, bad proportions, bad hands, deformed, mutated, loli, shota, aged down, child, extra digits, fewer digits, 6_fingers, 4_fingers, 1_finger, 2_fingers, 3_fingers, 3D, semirealistic, photorealistic, realistic</code></pre>Ddded lineart to some of the images but in negative prompt.Training Params:<h2 id="v0.8-version-information:">V0.8 Version Information:</h2><ul><li>Dataset includes images made from the previous datasets and cleaned it up.</li><li>Dataset includes images from only the LoRA mix used to make the beta version.</li><li>Dataset also includes images using V0.5</li><li>Dataset now includes images of Nano Banana to hopefully improve it.</li></ul><h3 id="for-those-who-it-may-concern.">For those who it may concern.</h3>Training Params:{ "engine": "kohya", "unetLR": 0.0001, "clipSkip": 2, "loraType": "lora", "keepTokens": 1, "networkDim": 64, "numRepeats": 1, "resolution": 1024, "lrScheduler": "cosine_with_restarts", "minSnrGamma": 5, "noiseOffset": 0.1, "targetSteps": 1425, "enableBucket": true, "networkAlpha": 32, "optimizerType": "Adafactor", "textEncoderLR": 0.00005, "maxTrainEpochs": 10, "shuffleCaption": true, "trainBatchSize": 4, "flipAugmentation": false, "lrSchedulerNumCycles": 3 }<h2 id="v0.5-version-information:">V0.5 Version Information:</h2><ul><li>Dataset includes images made from the previous datasets and cleaned it up to just the images I want for the style</li><li>Dataset includes images from only the LoRA mix used to make the beta version.</li><li>Dataset also includes images using the LoRA mix from the Mix I used with Beta plus V0.1 of the LoRA added into said mix.</li></ul><h3 id="for-those-who-it-may-concern.">For those who it may concern.</h3>The checkpoint used to generate images in the dataset: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2188538?modelVersionId=2464202">Hex Furry and More V0.5</a> since it likes the LoRA somewhat<a target="_blank" rel="ugc" href="https://civarchive.com/models/1595884?modelVersionId=1805971"> </a>also used <a target="_blank" rel="ugc" href="http://civarchive.com/models/2179383?modelVersionId=1805971">Crystal VAE V3</a>.And these for the prompts and settings:Sampler EulerScheduler: NormalSteps: 20CFG: 4Positive:<pre><code>[PROMPT], illustration, anime coloring, 2GDA1, zzkr_9528, N0xfallRev, animalshapes, ep1c7, arctenox, r1ng3k0, masterpiece, best quality, highest quality, absurdres, highres, extremely detailed, mature, official art, official style, [(artist:nt00:0.01), (artist:fuzichoco:0.01), (artist:swav:0.03), (artist:crote:0.03), (artist:unfairr:0.04), (fate_/(series/):0.2), (bleach:0.5), (black_clover:0.3)], usnr, Flatline, Flat vector illustration, 0kaz3, shi3j_illu, e0ce69, skyger_style</code></pre>Negative:<pre><code>worst quality, low quality, normal quality, lowres, blurry, out of focus, jpeg artifacts, compression artifacts, ugly, bad anatomy, bad proportions, bad hands, deformed, mutated, loli, shota, aged down, child</code></pre>Training Params: { "engine": "kohya", "unetLR": 0.0001, "clipSkip": 2, "loraType": "lora", "keepTokens": 1, "networkDim": 32, "numRepeats": 1, "resolution": 1024, "lrScheduler": "cosine_with_restarts", "minSnrGamma": 5, "noiseOffset": 0.1, "targetSteps": 1050, "enableBucket": true, "networkAlpha": 16, "optimizerType": "Adafactor", "textEncoderLR": 0.00005, "maxTrainEpochs": 10, "shuffleCaption": true, "trainBatchSize": 4, "flipAugmentation": false, "lrSchedulerNumCycles": 3 }<h2 id="v0.1-version-information:">V0.1 Version Information:</h2><ul><li>Dataset Includes images made from only the Beta Version.</li><li>Dataset includes images made from the Beta Version in a Small LoRA Style Mix.</li><li>Dataset includes images from the dataset that made the Beta Version.</li><li>Dataset includes a little bit of images I fixed through Nano Banana.</li><li>Dataset includes images from the LoRA Mix that made the Beta Version. </li></ul><h3 id="for-those-who-it-may-concern.">For those who it may concern.</h3>The checkpoint used to generate images in the dataset: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2307532?modelVersionId=2596271">Hex Ionsyx V1.0</a>, and <a target="_blank" rel="ugc" href="https://civarchive.com/models/2338192/atlas-core">Atlas Core V0.6</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/1595884?modelVersionId=1805971">Hyphoria V0.01 </a>also used <a target="_blank" rel="ugc" href="http://civarchive.com/models/2179383?modelVersionId=1805971">Crystal VAE V3</a>.This LoRA Mix in question uses:<ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2390241">Arctenox's Styles - Ouroboros | Series 2 [Illustrious]</a> on 58%</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2392019/lobsteranian-style-illustrious?modelVersionId=2689549">Lobsteranian style - Illustrious</a> on 11%</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/448943/bleach-anime-style-lora">Bleach Anime Style LoRA</a> on 10%</li></ul>And these for the prompts and settings:Sampler: Euler Ancestral / EulerScheduler: NormalSteps: 22 / 20 / 25CFG: 3.7 / 4Positive for the LoRA Mix made images inside of the Dataset:<pre><code>illustration, anime, 2390241, [PROMPT], masterpiece, best quality, highest quality, absurdres, highres, extremely detailed, mature, official art, official style, [(artist:nt00:0.01), (artist:fuzichoco:0.01), (artist:swav:0.03), (artist:crote:0.03), (artist:unfairr:0.04), (fate_/(series/):0.2), (black_clover:0.3)], bleach_tybw_style, lobsteranian</code></pre>Positive for the Beta Version of the LoRA only made images inside of the Dataset: A few images use this:<pre><code>illustration, anime, 2390241, masterpiece, best quality, highest quality, absurdres, highres, extremely detailed, mature, official art, official style, [PROMPT]</code></pre>The rest use this:<pre><code>[PROMPT] 2390241, masterpiece, best quality, highest quality, absurdres, highres, extremely detailed, mature, official art, official style, [(artist:nt00:0.01), (artist:fuzichoco:0.01), (artist:swav:0.03), (artist:crote:0.03), (artist:unfairr:0.04)]</code></pre>Negative used for everything:<pre><code>worst quality, low quality, normal quality, lowres, blurry, out of focus, jpeg artifacts, compression artifacts, ugly, bad anatomy, bad proportions, bad hands, deformed, mutated, loli, shota, aged down, child, extra digits, fewer digits, 6 fingers, 4 fingers, 3D, semirealistic, photorealistic, realistic</code></pre>Training Params: { "engine": "kohya", "unetLR": 0.0001, "clipSkip": 2, "loraType": "lora", "keepTokens": 1, "networkDim": 64, "numRepeats": 1, "resolution": 1024, "lrScheduler": "cosine_with_restarts", "minSnrGamma": 5, "noiseOffset": 0.1, "targetSteps": 925, "enableBucket": true, "networkAlpha": 32, "optimizerType": "AdamW8Bit", "textEncoderLR": 0.00003, "maxTrainEpochs": 10, "shuffleCaption": true, "trainBatchSize": 4, "flipAugmentation": false, "lrSchedulerNumCycles": 3 }<h2 id="beta-version-information:">Beta Version Information:</h2>This is a style mix made using:<ul><li>Dataset includes images using a LoRA mix.</li></ul><h3 id="for-those-who-it-may-concern.">For those who it may concern.</h3>The checkpoint used to generate images in the dataset: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2307532?modelVersionId=2596271">Hex Ionsyx V1.0</a>The LoRA Mix in question uses:<ol><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2387976?modelVersionId=2685156">Ringeko | by UTdT69</a> on 78%</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2330394/epic-seven-official-art-style-or-series-2-illustrious">Epic Seven - Official Art Style | Series 2 [Illustrious]</a> on 62%</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/861014?modelVersionId=1397472">Animal Shapes Style</a> on 50%</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/1254852/zackary911-style-illustrious?modelVersionId=1414712">Zackary911 Style illustrious</a> on 50%</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2179404/arctenoxs-styles-my-style-or-series-1-illustrious">Arctenox's Styles - My Style | Series 1 [Illustrious]</a> on 37%</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/1233370?modelVersionId=1805073">Flatline - Illus x Niji [BETA]</a> on 32%</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2203258/arctenoxs-styles-noxfall-or-series-1-illustrious">Arctenox's Styles - Noxfall | Series 1 [Illustrious]</a> on 30%</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2179549/arctenoxs-styles-celfade-or-series-1-illustrious">Arctenox's Styles - Celfade | Series 1 [Illustrious]</a> on 28%</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2372545/okaze-or-by-utdt69">Okaze | by UTdT69</a> on 22%</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2202936?modelVersionId=2480487">Arctenox's Styles - Detailed Anime | Series 1 [Illustrious]</a> on 10%</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/981656/i-cant-believe-is-not-kegant?modelVersionId=2418683">I can't believe is not Kegant</a> on 5%</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/176554/usnr-style">薄塗り / USNR STYLE</a> on 2%</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2272035/shiej-or-shiiros-styles?modelVersionId=2557366">Shiej | Shiiro's Styles</a> on %</li></ol>And these for the prompts and settings:Sampler EulerScheduler: NormalSteps: 20 - 35CFG: 4Positive:<pre><code>[PROMPT], illustration, anime coloring, 2GDA1, zzkr_9528, N0xfallRev, animalshapes, ep1c7, arctenox, r1ng3k0, masterpiece, best quality, highest quality, absurdres, highres, extremely detailed, mature, official art, official style, [(artist:nt00:0.01), (artist:fuzichoco:0.01), (artist:swav:0.03), (artist:crote:0.03), (artist:unfairr:0.04), (fate_/(series/):0.2), (bleach:0.5), (black_clover:0.3)], usnr, Flatline, Flat vector illustration, 0kaz3, shi3j_illu, e0ce69, skyger_style</code></pre>Negative:<pre><code>worst quality, low quality, normal quality, lowres, blurry, out of focus, jpeg artifacts, compression artifacts, ugly, bad anatomy, bad proportions, bad hands, deformed, mutated, loli, shota, aged down, child</code></pre>Training Params: { "engine": "kohya", "unetLR": 0.0001, "clipSkip": 2, "loraType": "lora", "keepTokens": 0, "networkDim": 32, "numRepeats": 1, "resolution": 1024, "lrScheduler": "cosine_with_restarts", "minSnrGamma": 5, "noiseOffset": 0.1, "targetSteps": 555, "enableBucket": true, "networkAlpha": 16, "optimizerType": "AdamW8Bit", "textEncoderLR": 0.00005, "maxTrainEpochs": 10, "shuffleCaption": true, "trainBatchSize": 4, "flipAugmentation": false, "lrSchedulerNumCycles": 3 }

Arctenox's Styles - Ouroboros | Series 2 [Illustrious]

Less overfitted I think...also 570 img dataset I think. Probably works better on different checkpoints than what was used in the thumbnail plus different settings. I have no idea. But I do know it's at a point I can personally generate with and make a dataset with primarily images made from this lora and maybe a few from this dataset for this LoRA's V1.0.

Files with SHA256: d3142607b40cc60aa37c87f9a7568dc489a4ecfaa35904741e6aabc6af16d5ca