Trained on Illustrious-XL.
ddkk:
Interestingly, I trained a style LoRA using a dataset based on this X user's works – the catch is that his art is also AI-generated. Despite this, I still named the style after them. Pretty meta, right?
Prompt :
masterpiece, best quality, amazing quality, Negative Prompt :
bad quality, worst quality, worst detail, sketch, censor, optional negative prompt :
monochrome, greyscale, signature, url, artist name, username, text, English text,Description
{
"engine": "kohya",
"unetLR": 1,
"clipSkip": 2,
"loraType": "lora",
"keepTokens": 1,
"networkDim": 16,
"numRepeats": 7,
"resolution": 1984,
"lrScheduler": "cosine",
"minSnrGamma": 5,
"noiseOffset": 0.1,
"targetSteps": 3945,
"enableBucket": true,
"networkAlpha": 8,
"optimizerType": "Prodigy",
"textEncoderLR": 1,
"maxTrainEpochs": 46,
"shuffleCaption": true,
"trainBatchSize": 4,
"flipAugmentation": false,
"lrSchedulerNumCycles": 3
}
FAQ
Comments (4)
When everyone is still guessing, trying to modify the prompt words or using the existing LoRa to reproduce the style of an AI artist who does not disclose metadata, choosing to directly convert AI images into LoRa is undoubtedly the most efficient choice. Suddenly I feel that this approach is very punk, very Gigachad......
we just need to identify wich artists they use tho
I believe DDKK's art can be trained into high-quality gut-torturing concept LORA too... So many gut-punching pics made.
Not bad, it need add note use better 0.2 than 1 try fix different but hard time with character lora
Great work :)