There is absolutely no reason for this LoRA to exist, but the SD 1.5 version is inexplicably one of my most popular LoRAs on CivitAI, so I thought I might as well make an SDXL version of it.
Please don't hate me. 😅
Description
FAQ
Comments (3)
Thank you for this model! For some reason combining with my own LoRA's it gives much higher quality output - less artefacts and not grainy/blurry at all. But also preserving my models effects and not only for on/offs. Usually I have to play with prompts a lot to achieve relatively good results. Very nice side-effect. What training resolution was used or maybe the secret is in lots of epochs and less steps per epoch?
Nice! I'm pretty sure it was 232 images, at 1024x1024 resolution, 2 repeats, batch size 2 for 12 epochs. I can double check, though
@Invisidude thank you! I never tried yet such process when you set small repeats number along with many epochs. Usually I do 2-3 epochs and 20-40 repeats. I read some articles which said more epochs and less steps is more preferable for getting a good output, but for some reason I ignored it. Of course there are other aspects, but maybe it can improve a bit my own results in some cases. Thank you again - both for this model and for your reply.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.