32 NetDim LoRAs trained on 365 watercolor painting images.
FluxDim32clipSkip1_nocap was trained with 1024x1024 images with no captions.
The following models were trained with captions prepended by "A wtrclr watercolor painting of ". Many of the images were also appended by "in the style of a wtrclr watercolor painting.":
FluxND32CS2_512_wtclr: Trained with 512x512 images and Clip Skip 2.
FluxND32CS1_1024_wtclr: Trained with 1024x1024 and Clip Skip 1.
Description
Model selected at epoch 12 of 20 (step 5840)
Training Params:
{
"engine": "kohya",
"unetLR": 0.0009,
"clipSkip": 2,
"loraType": "lora",
"keepTokens": 0,
"networkDim": 32,
"numRepeats": 4,
"resolution": 512,
"lrScheduler": "cosine",
"minSnrGamma": 5,
"noiseOffset": 0.1,
"targetSteps": 9734
"enableBucket": true,
"networkAlpha": 16,
"optimizerType": "Adafactor",
"textEncoderLR": 0,
"maxTrainEpochs": 20,
"shuffleCaption": false,
"trainBatchSize": 3,
"flipAugmentation": true,
"lrSchedulerNumCycles": 1
}