A LoRA for recreating Touka Kirishima character. The model also has a grasp of the Ken Kaneki's character. Works with dev, <a rel="ugc" href="https://civarchive.com/models/756183/raemu-flux">raemu </a>and <a rel="ugc" href="https://civarchive.com/models/1071160/unbelievable">unbelievable_v10 </a>models.

Touka Kirishima (Tokyo Ghoul) [Flux dev]

Optimized training params a bit: {
 "engine": "kohya",
 "unetLR": 0.00025,
 "clipSkip": 1,
 "loraType": "lora",
 "keepTokens": 0,
 "networkDim": 32,
 "numRepeats": 6,
 "resolution": 1024,
 "lrScheduler": "cosine_with_restarts",
 "minSnrGamma": 5,
 "noiseOffset": 0.1,
 "targetSteps": 2040,
 "enableBucket": true,
 "networkAlpha": 16,
 "optimizerType": "AdamW8Bit",
 "textEncoderLR": 0,
 "maxTrainEpochs": 20,
 "shuffleCaption": false,
 "trainBatchSize": 2,
 "flipAugmentation": false,
 "lrSchedulerNumCycles": 3
}

touka_kirishima_flux.safetensors

v1.0

A LoRA for recreating Touka Kirishima character. The model also has a grasp of the Ken Kaneki's character. Works with dev, <a href="https://civitai.com/models/756183/raemu-flux" rel="ugc noopener noreferrer nofollow" target="_blank">raemu </a>and <a href="https://civitai.com/models/1071160/unbelievable" rel="ugc noopener noreferrer nofollow" target="_blank">unbelievable_v10 </a>models.

&lt;p&gt;Optimized training params a bit:&lt;br /&gt;{
  &#34;engine&#34;: &#34;kohya&#34;,
  &#34;unetLR&#34;: 0.00025,
  &#34;clipSkip&#34;: 1,
  &#34;loraType&#34;: &#34;lora&#34;,
  &#34;keepTokens&#34;: 0,
  &#34;networkDim&#34;: 32,
  &#34;numRepeats&#34;: 6,
  &#34;resolution&#34;: 1024,
  &#34;lrScheduler&#34;: &#34;cosine_with_restarts&#34;,
  &#34;minSnrGamma&#34;: 5,
  &#34;noiseOffset&#34;: 0.1,
  &#34;targetSteps&#34;: 2040,
  &#34;enableBucket&#34;: true,
  &#34;networkAlpha&#34;: 16,
  &#34;optimizerType&#34;: &#34;AdamW8Bit&#34;,
  &#34;textEncoderLR&#34;: 0,
  &#34;maxTrainEpochs&#34;: 20,
  &#34;shuffleCaption&#34;: false,
  &#34;trainBatchSize&#34;: 2,
  &#34;flipAugmentation&#34;: false,
  &#34;lrSchedulerNumCycles&#34;: 3
}&lt;/p&gt;

Files with SHA256: d659dbe8263f04b13addf129aebe4deef65d8aa3306eec7cc95a9128598c9d19