Support my work on Patreon and Ko-Fi and get access to tutorials and exclusive models.
Rate and leave a like if you enjoyed it, and follow for new models!
I did this LoRA because I wanted to test doing a LoRA for a small object with little impact on the image (see examples below). I tried doing a LyCORIS but still haven't found the ideal parameters.
I recommend using 0.5-0.8 weight, but you can use 1.0 without impacting the style of the image.
You can use with glasses-related tags, but it works better when using: circle-glasses
(I do realize there is another LoRA for round glasses out there but it wasn't working for me because, in my experiments, it affects style and the composition too much.)
Weight examples and effect on image:
Description
44 images, 10 repeats, 10 epochs, batch size 6
LR 0.0001, AdamW8bit
512,512 resolution
8 network dim/alpha
(use Additional Networks in webui to view more metadata for this LoRA)
FAQ
Comments (7)
很棒~
Looks great! If you want to do LyCORIS you could try to extract that from a Dreambooth model way easier. I did a PDF guide on this a couple of days ago and get amazing results this way! Maybe you should give it a try? https://civitai.com/models/62914/easy-lycoris-extraction-from-trained-dreambooth-model
Thanks for sharing. I still need some time to work through that info.
But this is a lora. Is your idea basically to train dreambooth directly instead of a lora to then extract a lycoris?
@psoft yes exactly! Others have done the same and can confirm that the quality is much much better than directly train a LoRA. And it is also way easier.
@technerd Lora training is very similar, no? How’s it easier? And what about time?
@psoft read the PDF it's only 2 pages. Everything explained there, no caption textfiles needed. NMKD GUI for Dreambooth training via included Presets. With only the Medium Quality preset within 20 minutes I got fantastic results. LyCORIS extraction took only 20 seconds on my old i7-6700K system.
Hi, :( there is no Pony version of this beautifull lora ?
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.