Expand {
"engine": "kohya",
"unetLR": 0.0005,
"clipSkip": 1,
"loraType": "lora",
"keepTokens": 3,
"networkDim": 64,
"numRepeats": 4,
"resolution": 1024,
"lrScheduler": "cosine_with_restarts",
"minSnrGamma": 5,
"noiseOffset": 0.1,
"targetSteps": 9882,
"enableBucket": true,
"networkAlpha": 16,
"optimizerType": "Adafactor",
"textEncoderLR": 0.00005,
"maxTrainEpochs": 18,
"shuffleCaption": true,
"trainBatchSize": 4,
"flipAugmentation": true,
"lrSchedulerNumCycles": 3
}Description
i think i've been training wrong... too many images for the AI to learn from, so now i'm being more selective of the images generated by midjourney and adding them to the lora...
as usual, its old men with "equinecock" "horsecock"
generally use lower for all of my loras.
sample images are using the following:
[
{"prompt": "married malay man going to work"},
{"prompt": "married grandfather going to work"},
{"prompt": "grandfather with extremely large"}
]