Model examples:
Z-Turbo V2;
Z-Turbo- V1;
Klein 9B - V2;
Pony V3;
SDXL-V4;
SDXL-V3;
SDXL-V2;
SDXL-V1;
Description
New dataset practices, Should respond to different image sizes better.
Training settings: (this is for me)
"lr": 0.0005,
"engine": "ai-toolkit",
"epochs": 20,
"ecosystem": "sdxl",
"keepTokens": 1,
"networkDim": 32,
"numRepeats": 2,
"resolution": 1024,
"lrScheduler": "cosine",
"minSnrGamma": 5,
"noiseOffset": 0.03,
"networkAlpha": 32,
"optimizerType": "adamw8bit",
"shuffleTokens": true,
"textEncoderLr": 0.00005,
"flipAugmentation": false,
"trainTextEncoder": true