Trained on realistic images focus on POV.
best output is around 0,8-1 for me
Description
FAQ
Comments (12)
I'm trying to learn how to create concept LoRA for Flux. what was you dataset size? Did you use resolution 512 or 1024?
dataset containing around 50 images and resolution from 1024-2048
@mrkakapopoloch so the resolution of the training (civitai asks you to input a resolution) was set at 1024?
Or when you train locally you do not put that?
@barlogspam176 civitai asks you to input a MAX resolution, so answer is 2048
@acaciahoney Ah. thank you for the tip. didn't realize that
Does this work with the nf4 model? Have tried a few generations and can't get anything even similar to a blowjob.
for me it does work with nf4, yes
@mrkakapopoloch I got it working. I had to update Forge and change a setting.
@skullzy77 which setting?
@dchen19 Up at the top in the area that you choose your checkpoint there is an option called Diffusion In Low Bits. I changed that option to Automatic (fp16 LoRa)
Although visibly trained with female character, it works reasonably well with male characters at the price of a small distortion (like a loss of facial hair).
easily the best Flux blowjob that doesn't alter character lora faces. Keep up the fantastic work.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.