
Special thanks:
V2.0:
Reduced "networkAlpha" from 16 to 12, changed to Prodigy, and added training steps.
17 training pictures with natural language tags.
Epoch #23 (17×6×23=2,346 steps) seems to perform better than Epoch #25 (2,550 steps).
{
"engine": "kohya",
"unetLR": 1,
"clipSkip": 1,
"loraType": "lora",
"keepTokens": 0,
"networkDim": 2,
"numRepeats": 6,
"resolution": 512,
"lrScheduler": "cosine",
"minSnrGamma": 5,
"noiseOffset": 0.1,
"targetSteps": 850,
"enableBucket": true,
"networkAlpha": 12,
"optimizerType": "Prodigy",
"textEncoderLR": 0,
"maxTrainEpochs": 25,
"shuffleCaption": false,
"trainBatchSize": 3,
"flipAugmentation": false,
"lrSchedulerNumCycles": 3
}
Adjusting legwear (tried flux version).
You need to add some specific types of legwear to get good results (e.g., white socks).
Can be used with realistic lora.
描述:
训练词语: adjusting legwear
名称: 734271_training_data.zip
大小 (KB): 5560
类型: Training Data
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success
名称: adjusting_legwear_flux_v2.safetensors
大小 (KB): 18815
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success