If you'd like to support me and do more:
https://ko-fi.com/edgonkofi - commissions open!
If you're looking for a >>LoRA Making Tutorial<<
Hoodies - LoRA version
Please also check this model in order to compare: https://civarchive.com/models/19491/hoodies-first-edg-locon
Since it's my first time training LoCon, I thought I would release a LoRA version in case you could like to compare.
Please give me feedback!
Get your hoodies now!
Description
FAQ
Comments (9)
Nice! Wouldn't you know that the gen that's running right now... I started to put the person in a hoodie. Does it do solid color hoodies?
Hello dita
Solid color hoodies can be generated without LORA
This is for more intricate design
@EDG lol thanks. I actually had not even tried it yet. Probably should've done that before I asked, huh? 😋
@dita No worries, I actually am glad that you rely on me.
I discovered that native SD did some basic clothing very well when I tried to train for leather jackets.
Base SD is excellent with leather jackets already!
@EDG I do rely on you for sure!
I'm currently doing a run using the dynamic prompt extension with the tag ({1-2$$wearing leather thigh boots|wearing leather chaps|wearing leather bustier|wearing leather corset|wearing leather thong}:1.15) (i.e., select 1 or 2 of each of these options, in case you aren't familiar with the syntax)
I've done this one before, and it generally does quite well with those items as well. It's definitely impressive in some ways. And sometimes it scares me when I use a prose-type of prompt just how much better it turns out than when I use a list of tags—sometimes even with a model trained on tags.
@dita you're right, even without LoRA you can get great things
This is what pushes me to make good models
If someone can reach your LoRA only with a prompt then you should get back training!
I'm very inspired by good prompts and I strive to create content that gives good results to novice promoters while allowing master promoters to reach new heights!
That's my purpose
@EDG I like that philosophy. I think that you're using the technology in the best way possible. You are maximizing the benefit for the most number of users, focusing on areas where there has been little content created previously, and not creating redundant, unwieldy, difficult, or unneeded things. I don't know how many times I've seen multiple LoRAs come out for the same character, and I have to think it's a waste of resources.
Right now it's easy to keep making LoRA, LoCons, LoHas, and whatever else that have a lot of value; the field is wide open. If I sat down and made a list, I could probably come up with 100 to 200 LoRAs that I want—just in terms of anime characters. Eventually it will become more niche. I think we're already seeing that with checkpoints, now that both style and content can be determined almost entirely by one or more LoRAs. Most checkpoints these days are just a remix of the same dozen or so base checkpoints, in different combinations and arrangements. Maybe one day, we'll be at the point where everyone uses only one checkpoint for basic functionality, and all the content and style are determined by the LoRAs they use.
@dita Thank you for sharing your vision on the present and future of this technology.
Let's stick around and try to move things in the right direction!
@EDG That's my plan. :) Probably more as a consumer and commentator than an innovator, but I'll leave that part to you. 😄
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.