Way Better Then Dreambooth! Custom AI Models Based on Your Photos!
Vložit
- čas přidán 8. 06. 2024
- Perfusion lets you customize AI image generation with your own concepts using compact model updates! See how locking attention keys and training latent values enable quality text-to-image personalization.
▼ Link(s) From Today’s Video:
✩ Nvidia's Paper: arxiv.org/pdf/2305.01644.pdf
► MattVidPro Discord: / discord
► Follow Me on Twitter: / mattvidpro
-------------------------------------------------
▼ Extra Links of Interest:
✩ AI LINKS MASTER LIST: www.futurepedia.io/
✩ General AI Playlist: • General MattVidPro AI ...
✩ AI I use to edit videos: www.descript.com/?lmref=nA4fDg
✩ Second Channel: / @matt_pie
-------------------------------------------------
Thanks for watching Matt Video Productions! I make all sorts of videos here on CZcams! Technology, Tutorials, and Reviews! Enjoy Your stay here, and subscribe!
All Suggestions, Thoughts And Comments Are Greatly Appreciated… Because I Actually Read Them.
-------------------------------------------------
► Business Contact: MattVidProSecond@gmail.com - Věda a technologie
Ah, the 'it exists but you can't use it,' genre of AI news. The corner stone of frustration.
I only really call bs when it's Google though.
@@virgilxavier1 omg those guys. They told us for years,”You don’t deserve this”
As long as they explained how it's done so we can replicate
@@timmygilbert4102 ahhh great point
I wish people would say that up front. This video is a waste of time
This reminds a lot of the character ai website, but with images. Can't wait for this model to be usable (and hopefully it's free)
This is some cool stuff man. Always excited to see your videos in my notifications inbox.
Agreed. Matt is the best. Always entertaining.
I am genuinely confused why they don't just release this already?
Why should they? Don't think they let people like us profit from that. They first are going to make tons of money with most of their Technology ofc.
It's like people saying people using AI will replace other people but no that's very wrong. People owning the AI will replace people who don't own one 😂
@@David.Alberg I get that for things like language models, but i don't really get how not releasing this could be beneficial.
@@RandumbThings it's all about capitalism and competition. The world runs on money ofc. I really hope that in the next 5 years we are not dependent on money and everything will be much better.
Because it's not 2030/40 yet.
Around that time it'll be less about money and more about status
@@Gerlaffy you are right but I guess you are underestimating exponential growth so it will be probably earlier
I feel like this is the biggest AI development in months. AI takes mountains to get a single concept. A human can see a single picture and learn the concepts from it based on inferences from their life experience. To create such a small file on a concept from a single example means AI is getting smart-scary smart.
They do have their database to combine/ train with the pictures.
dude i remember playing with you and dirtbike and rcsandy (or something like that) wayyy back in the day, youre channel has grown so much! you might remember me, i was AkayAllie
damn, cant wait for the model to release
Hey Chat GPT is the title of this video gramatically correct?
No, the sentence "Way Better Then Dreambooth! Custom AI Models Based on Your Photos!" is not grammatically correct. The word "then" should be "than" when making comparisons. The corrected sentence would be: "Way Better Than Dreambooth! Custom AI Models Based on Your Photos!"
You really need ChatGPT for that? Damn humanity is doomed
Hold onto your papers, fellow scholars...
real
Clean Cut Matt
How is it different than LoRAs, just smaller size?
filesize is one thing, but it seems like the big perk is being able to use a ton of these fine-tuned model extensions together and also that they are much more coherent than LoRA. In my experience, LoRA does great for styles and pretty good on characters, but when you throw in a new concept they don't often play nice with each other. This new technique seems like it can keep things much more coherent if you're pumping a ton of concepts, styles, and characters into the model.
is this available yet? and or is there something else out there now?
FYI there's a typo in your description, it says "then" instead of "than" 👍
At least we know he isnt using ai to generate his titles 😅
Stable diffusion textual inversions do pretty much the same thing and are small in file size.
But they are not as coherent as this method. Textual inversion is more a concept of styles than subjects
Just release the thing and Matt can put the link up!
Very nice
3:36 Gotcha, Matt!!
oh god cant wait
Any update on this?
What will Matts next video might be hmm🤔? Oh i think it will be on "FullJourney" am i right???😄
make a video on firefly generative font styles
So the most important question is.......when will the public get access to NVidia's model?
never, but we have free AI models, SD1.5 or the new SDXL, that's all you really need.
Wish we can use this one day!
based
Damn its like textual inversion on crack
I like your shirt.
GRAMMAR ALERT!!!
YOU PUT THEN, NOT THAN!!!
pls fix ty :)
gotta wait for someone smart to make it into an a1111 extension
Assuming that the chosen examples shown are NOT cherry-picked is incredibly naïve.
3:36 😂
*than
1h 137 likes 207k suns
it's "than", not "then"
Create a new pink Floyd music with AI please
a
the
fox
jumps
over
the
lazy
'Super impressive stuff". Really? Its hardly any better in the examples you illustrate.