This is a furry/anthro gay male barazoku LORA based on Zhaotorama. This used the SDXL Model, Standard, and was trained on 166 images. To note, there is a lot of Bowser, Tigers, Lions, and Wolves that this was trained on, with a smattering of other characters.
Feel free to offer advice or tweaks! This is my first attempt at a LoRA, and I assume I'll be looking to try again.
TRAINING PARAMETERS
"engine": "kohya",
"unetLR": 0.0005,
"clipSkip": 1,
"loraType": "lora",
"keepTokens": 1,
"networkDim": 32,
"numRepeats": 3,
"resolution": 1024,
"lrScheduler": "cosine_with_restarts",
"minSnrGamma": 5,
"noiseOffset": 0.1,
"targetSteps": 1494,
"enableBucket": true,
"networkAlpha": 16,
"optimizerType": "Adafactor",
"textEncoderLR": 0.00005,
"maxTrainEpochs": 12,
"shuffleCaption": true,
"trainBatchSize": 4,
"flipAugmentation": true,
"lrSchedulerNumCycles": 3Description
This is not only the very first version of this LoRA, it's my very first LoRA period! Any advice or tips would be appreciated.