how does it look like with pixart dreambooth training? any news?
how does it look like with pixart dreambooth training? any news?
Fstill didnt have chance to try yet
Sthanks, looking forward to it
Mim looking for kaggle notebook for kokhya anyone can help me?
Fyes it is available on our patreon
F
DAnybody here has a good tech for lips synch? I use wave 2 lips and it s really bad most of the time and really blured.
Fi think there arent public tech yet
Fbetter
DAh ok, i ve been trying them all, but i was desperatly lookunh for the kind of ones used for heygen for instance.
Fprobably they are hidden or paid
DYeah! I guess
Lwhat you guys think of my new lawyer logo? i stole it from Better Call Saul, can't help but feel like something is missing though.
what would you guys have done instead? or to improve it
what would you guys have done instead? or to improve it
Lworld's first AI attorney hopefully
Fare you a lawyer?
X@Furkan Gözükara SECourses What do you think about Air.ai? will open ai be distroying it like it did with Jasper.ai?
LThey might end up taking themselves out 
I listened to their demo and imagined I was the customer, if I found out I was "joking" with an AI and chatting about Musk and fights .. guess my mood about Apple ?
I listened to their demo and imagined I was the customer, if I found out I was "joking" with an AI and chatting about Musk and fights .. guess my mood about Apple ?
XBut the screenshot you sent is more of a text conversational, AI
XAir ai is a “voice”
LThat voice has a normal language model behind it
LBut the demo is well done, it even included tts errors. so it's probably quite honest in capabilities.
However it was a perfect use case example, normal people are hard to predict.
In any case it's more for illegal spam callcenters than for businesses.
If it doesn't tell me right away that it is an AI I will be super pissed once I find out.
If it tells me it is an AI I'll hang up
However it was a perfect use case example, normal people are hard to predict.
In any case it's more for illegal spam callcenters than for businesses.
If it doesn't tell me right away that it is an AI I will be super pissed once I find out.
If it tells me it is an AI I'll hang up
XI’ve been trying to automate a call center
LIt's the future of callcenters - just that we might not need people or centers anymore at one point.
Right now - everyone will absolutely hate AI calls. In EU it's illegal to use AI to "influence" people so they will probably just ban calls (and ban anything else with AI sooner or later).
In maybe 5 years you might actually want AI, especially for support. Because it's smarter and without waiting time until you need a transfer to a human supervisor.
In maybe 15-20 years you'll not find a human anywhere in that chain
Right now - everyone will absolutely hate AI calls. In EU it's illegal to use AI to "influence" people so they will probably just ban calls (and ban anything else with AI sooner or later).
In maybe 5 years you might actually want AI, especially for support. Because it's smarter and without waiting time until you need a transfer to a human supervisor.
In maybe 15-20 years you'll not find a human anywhere in that chain
LI'm heading to bed, gnight
XSucks, but I agree ai has a huge side eye from a lot of call center kind of businesses. However, I don't think we'll see call center employees reciting the same script over and over again in the next two years.
XSo overall conversational AI for scheduling appointments isn’t really good yet, you think?
XThis is very dangerous lol!!!
LOh I think it definitely is good enough but most people will have problems implementing something that does the job sufficiently. It won't work out of the box.
The image I posted is just a good example what happens if you put someone incompetent behind it.
Models, finetunes and commercial APIs are all improving at a pace that is almost unseen in human technological progress, so it will get easier
The image I posted is just a good example what happens if you put someone incompetent behind it.
Models, finetunes and commercial APIs are all improving at a pace that is almost unseen in human technological progress, so it will get easier
Fit will cause some unemployment certainly
Fby the way
Fif i i understand ai talking i dont talk anymore
XWhat do you mean?
Ac
Fi mean in customer services i hate AI 
MYouTubeSECourses
Master Stable Diffusion XL Training on Kaggle for Free! Welcome to this comprehensive tutorial where I'll be guiding you through the exciting world of setting up and training Stable Diffusion XL (SDXL) with Kohya on a free Kaggle account. This video is your one-stop resource for learning everything from initiating a Kaggle session with dual ...
Master Stable Diffusion XL Training on Kaggle for Free! Welcome to this comprehensive tutorial where I'll be guiding you through the exciting world of setting up and training Stable Diffusion XL (SDXL) with Kohya on a free Kaggle account. This video is your one-stop resource for learning everything from initiating a Kaggle session with dual ...Mand does lora need regularization images? if so which set would you recommend for a style lora?
Ahopefully we will get a dreambooth guide for pixart soon, i got it working, but im still a bit uncertain on the parameters.
also is it possible to get around the 120 token truncating with pixart?
also is it possible to get around the 120 token truncating with pixart?
Fyou can
Ffor style you usually dont need
Fjust have good and big dataset
Fi never had chance to try yet sadly
Mfor style dreambooth training (to later extract lora) can i use 10-15 epoch and lesser repeats like 25? also save every n steps at 0?
Mbecause in your training video for faces you set epoch to 1 and repeats to 100 i was wondering if my settings would be suitable for style if i keep everything the same in your v12 json but just change those settings i mentioned above
Mand one last thing because the vae is set to stabilityai/sdxl-vae in the v12 json file, when using the trained lora later is it going to be a problem if i use it with a checkpoint that needs a vae? should i delete that setting if i'm going to extract lora from the dreambooth training later?
Fdepends on number of training images
Fit shouldnt matter much
M"I recommed tweaking your repeats so that a single epoch (number of images × number repeats) is 200-400 steps." is this a good formula?
Mso 200-400 divided by the number of training images we have?