Trained on the CivitAI trainer
Same dataset as the SD1.5 version.
Adding some important training notes:
I tried making this with 4 repeats and 2 batch (cost 688 buzz) but it failed.
The version that worked was made with with 6 repeats and 2 batch (cost 906 buzz)
Full list of settings
unetLR: 0.0005
clipSkip: 1
loraType: lora
keepTokens: 1
networkDim: 32
numRepeats: 6
resolution: 1024
lrScheduler: cosine
minSnrGamma: 0
noiseOffset: 0.03
targetSteps: 2190
enableBucket: true
networkAlpha: 32
optimizerType: Prodigy
textEncoderLR: 0.00005
maxTrainEpochs: 10
shuffleCaption: true
trainBatchSize: 2
flipAugmentation: false
lrSchedulerNumCycles: 3










