My first attempt at a LORA so don't expect too much... but I think the LOAR is quite usable.

PLEASE SHARE YOUR RESULTS :-)

... and please leave me a like, if you like :-) it would be important for me to see what is possible with the LORA. Thank you :-)

I used the LORA with flux1-dev-fp8 - Euler/Simple - 30 steps on ComfyUI.

The optimal LORA strength for me is between 0.7 and 0.8. Just play around a bit.

TRIGGER: dild0, penisdild0, d0ubledild0, 4nalp1ug, st00ldil0, vibrat0r, dild0lick1ng, b00tledild0

you may use my workflow with SD15 Upscale for realistic pictures

I created the LORA with CivitAI-Training. I used about 80 images with a resolution of 1024x1024.

Ressources:

SD15 Upscale model AngrA RealFlex close to real

Upscaler 4x_NMKD-Siax_200k

flux realism lora

I would be very happy if you would leave me suggestions for the generation of LORAs in the comments.

I have no idea which settings make sense for taring... I have set "something" and would be happy to receive suggestions for optimisation ;-)

{

"engine": "kohya",

"unetLR": 0.0005,

"clipSkip": 1,

"loraType": "lora",

"keepTokens": 0,

"networkDim": 50,

"numRepeats": 46,

"resolution": 1024,

"lrScheduler": "cosine_with_restarts",

"minSnrGamma": 5,

"noiseOffset": 0.1,

"targetSteps": 9936,

"enableBucket": true,

"networkAlpha": 16,

"optimizerType": "AdamW8Bit",

"textEncoderLR": 0,

"maxTrainEpochs": 8,

"shuffleCaption": false,

"trainBatchSize": 3,

"flipAugmentation": false,

"lrSchedulerNumCycles": 3

}