This LoRA aids in generating images of subjects with wet clothes, particularly of women who are not wearing bras.
While it can work on colored clothing, it has the most effect on white clothes. The wet clothes keyword is optional but can have a big impact on the composition. (And since this LoRA did not train the text encoder, it's just the concept of wet clothing that triggers the effect, not the exact phrase.)
Description
FAQ
Comments (9)
Such a great idea.
Thanks!
And oh my gosh, I just noticed that you released a breasts-against-glass LoRA yesterday!
I just did the same... Great minds think alike! 😂
@Invisidude Your's is looking better than mine. I also made a wet shirt model lol.
@Shurik Great minds indeed! 😂 A wet T-shirt LoRA for SD1.5 was my first model on the site 👌
When it comes to Flux training, I'm finding that I get better results when I ignore some of CivitAI's current training recommendations. My best Flux outputs used over 100 training images and were trained at 1024x1024 resolution.
(I also have started making sure the number of repeats is equal to or divisible by my batch size in case there are images that aren't sized similarly enough to any others, in which case the repeats of such images can at least get lumped together as a batch instead of being excluded. That's not specific to Flux, just general bucketing.)
@Invisidude those seem like good ideas. I tried one 1024 and it didn't look that good to me. May have been an issue with the dataset though. I'll try it again sometime. I typically do 100+ images myself. I have been using repeat images rotated and zoomed to add variation. I feel like that works well for me. Unfortunately, flux is a little pricy for me to do a lot of experiments. I hate having to train a new version of a model.
@Invisidude can u tel me which model r u used? flux dev fp8 or gguf
@kalayanrao148252 flux dev fp8
You, sir, are a man of culture!
One of my favorite Flux Loras on the site. Powerful and easy to work with.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.