New Stable Diffusion Dreambooth Archive
HTML-код
- Опубликовано: 30 ноя 2022
- Civitai is an (awesome!) archive for stable diffusion dreambooth finetunes and it's really easy to work with. In this tutorial we walk through the process of downloading a model from Civitai and generating some images with it!
Discord: / discord
======= Links =======
The archive: civitai.com/
======= Music =======
Music from freetousemusic.com
‘Daily’ by ‘LuKremBo’: • (no copyright music) c...
‘Late Morning’ by ‘LuKremBo’: • (no copyright music) c...
‘Marshmallow’ by ‘LuKremBo’: • lukrembo - marshmallow...
‘Onion’ by LuKremBo: • (no copyright music) l...
#stablediffusion #aiart #tutorials #techtutorials #dreambooth #ai #machinelearning #tech 
Наука
Thanks for the shoutout! We're just getting started on this thing and It's exciting to see people talking about it.
Just wanted to clear one thing up, we actually do pickle scan everything that's shared. You can click on the shield next to the download button on every page to see the result of the scan. Our scanner is also open source so you can see exactly how we're scanning too :)
DUDE, that's so good!!!
I like where your mouse pointed in the very start of your video when you said "there's a big image"
HEY!! that was an accident I swear! Don't make me unleash muscular joe biden on you
Lots of dreambooth models require higher steps for good results ( 42-50 steps is my go to with majority of the custom trained models and I get the same results as the samples shown), having a cfg of 9-12 also really helps)
and some need high steps but super low cfg. you really need to experiment, lots of xy grids.
The inkpunk model might be overtrained per your examples, but it's easy to add randomness without excluding model influence. I noticed constrained variance until I added weights (style of nvinkpunk:0.4) and negative prompts, it "unfits" the model and gives great results!
That's super interesting, I'll have to try it out
Also, you misstated that the models are not scanned or verified... they definitely are, there's the green shield and checkmark to indicate that they are scanned for pickle imports. Just click on the shield in the model menu and you'll get the report.
it's cool to see my models on a youtube video!
thanks for sharing!
Ya i've tried trying my own hypernet models always turn out kinda crappy and my pc cant run dream booth local ☹ This is a good vid though gonna definitely check them out thanks !
It looks like the model you downloaded was based on stable diffusion 2.0 meaning its meant to generate 720 x 720 images (or whatever it actually is)
nice catch, I had downloaded so many models it was all starting to blur together
did you try using the models with img2img?
I tried a few of them, and the results seemed pretty normal... nothing crazy
WoooooO0O0OoooooW.. Tnx man!!
How come these models are all limited to 2GB size? Aren't they supposed to be trained on top of the original SD models of 4GB+?
It would really not hurt to have instructions on how to train in the new version. Automatic 1111 is now done there with a bunch of tabs, completely superfluous for me. And the main question is how to train for people, objects, and animals, this is understandable, but how to train the base for stylization?
ya nye gavaru pa russki 😢
@@lewingtonn one more
Not sure when it happened but a lot of these models have the new SafeTensor format which cannot include malicious code.
Finally
You should have used the Woolitize model for a more fun example ;)
Would advise after selecting one from here, just downloading from huggingface instead.
Afaik, those are scanned for viruses.
looks like the ones on the new site are scanned too, see the comment by the project owner above.
I noticed the textual inversion files just shows the images of the original artist/photos.
yeah, i realized that after too... has anyone had good experiences using TIV?
Yeah, we imported a bunch of Textual Inversions from HuggingFace and the only images they had were the source images. We're working on generating examples for all of those, but if anyone in the community is up for generating some examples and sharing them as reviews, that'd be appreciated :)
FJB!
is that a prompt? :P
ew biden
WHAT ID YOU JUST SAY????
@@lewingtonn exactly what you just read
but he's so funnnnnnyyyyyy
@@lewingtonn when He's not molesting toddlers sure. His stumbles and fumbles are certainly entertaining to watch
@@dannyd4339 molesting toddlers? damn that's a weapons grade take there haha. I don't judge. I see a man talk about hairy legs, I leave a like.