NOTE: DISCUSSION IS OPEN AGAIN ♥,

Trained this Lora on a personal style

Only trained on 79 personal images it can have shortcomings. choose the Epoch that's works for you. based on some reports I don't recommend strength 1.0 it will easily be overfitting. lower the value 0.75 - 0.50 - 0.30

♥ Realistic/3D Coming SOON! ♥

Training Params For who needs it

{
  "engine": "kohya",
  "unetLR": 0.0005,
  "clipSkip": 2,
  "loraType": "lora",
  "keepTokens": 0,
  "networkDim": 32,
  "numRepeats": 25,
  "resolution": 1024,
  "lrScheduler": "cosine_with_restarts",
  "minSnrGamma": 5,
  "noiseOffset": 0.1,
  "targetSteps": 2850,
  "enableBucket": true,
  "networkAlpha": 16,
  "optimizerType": "Adafactor",
  "textEncoderLR": 0.00005,
  "maxTrainEpochs": 12,
  "shuffleCaption": true,
  "trainBatchSize": 4,
  "flipAugmentation": true,
  "lrSchedulerNumCycles": 3
}

I decided to upload every Epoch for test purposes to find what works best without overfitting or undertraining

disclaimer: The Training images are personally generated by me

A Japanese Ai artist that uses (AI) to generate images (lmao) is angry that this model produces similar style. so, he and his fans attacking me and saying I stole his style!!! I'm not kidding x)you just can't make these things up