ZCIK is an artist I've been wanting to do for a super long time but I was worried their style wouldn't translate well into AI. Thankfully it seems I was wrong.
I recommend using a lot of prompting for lighting, like "flash photography" etc. As ZCIK tends to use stronger lighting in their higher quality images.
Regretfully there's no character tagging in the dataset as ZCIK's recurring characters don't appear often enough for that to be viable.
Description
Trained in 201 images:
{
"engine": "kohya",
"unetLR": 1,
"clipSkip": 2,
"loraType": "lora",
"keepTokens": 1,
"networkDim": 16,
"numRepeats": 3,
"resolution": 1024,
"lrScheduler": "cosine",
"minSnrGamma": 5,
"noiseOffset": 0.05,
"targetSteps": 3015,
"enableBucket": true,
"networkAlpha": 8,
"optimizerType": "Prodigy",
"textEncoderLR": 1,
"maxTrainEpochs": 20,
"shuffleCaption": true,
"trainBatchSize": 4,
"flipAugmentation": true,
"lrSchedulerNumCycles": 3
}
FAQ
Comments (3)
Hey, out of the other loras for this style, I like yours the most. Just need to ask which model/checkpoint do you recomend to use it? Some the images get the style some others I generated not so much but does work because of the faces.
@drarts90 Thanks for the support! I would first recommend checking the sample pics to see which look the best to you, they mostly use a wide variety of checkpoints for the sole purpose of giving people a showcase of how it looks with different models.
I'm very biased as I made 2 of the checkpoints there, but Indigo Genesis is built to be stable with this sort of flat color style in mind so it might be a good place to start if you haven't already tried that with this LoRA. The latest V17 of Wai is also pretty consistent and I used it for a few sample pics. Reed Furry Mix is also quite solid for this sort of style I find.
@Boerboer Thanks a lot!. I will try then share my new generations to see if I get right.