AI [AI Art] - Show Us Your AI Skill *NO TEENS*

4.80 star(s) 6 Votes
Discussion Reviews (6)
F

felldude

Active Member
Aug 26, 2017
572
1,724
cryptonaut said:
I don't like DALI or Midjourney and SD 1-1.2 leave much to be desired but 2+ works good once you have a well assembled model tamed by reasonable LoRA, a hypernetwork or two and a textual inversion, after that it's just a matter of dialing in the seed and tweeking the prompt. my solution can do very nice video too. i'll link a sample.

You must be registered to see the links


sorry for the rapid edits. theres a link to a sample
Click to expand...
With all the text to speech on the
You must be registered to see the links
Id like to get
You must be registered to see the links
working but maybe Ill apply for
You must be registered to see the links

It seems even Nvidia can't keep up with everything google and Facebook puts out.

Something weird I noticed is
You must be registered to see the links
uses CUDA CUDNN and CuBLAS (Over 2GB worth of .dll)
Its open beta and just thrown out there...I don't think the source is available it would be interesting to see how they are incorporating all of that.
 
Last edited:
C

cryptonaut

New Member
Sep 19, 2019
3
8
felldude said:
With all the text to speech on the
You must be registered to see the links
Id like to get
You must be registered to see the links
working but maybe Ill apply for
You must be registered to see the links

It seems even Nvidia can't keep up with everything google and Facebook puts out.

Something weird I noticed is
You must be registered to see the links
uses CUDA CUDNN and CuBLAS (Over 2GB worth of .dll)
Its open beta and just thrown out there...I don't think the source is available it would be interesting to see how they are incorporating all of that.
Click to expand...
As of today i have officially migrated completely away from fix point models, all my data point and tokenization is now dynamically stored in a graph db fully multimodal so now i have a unified API interface for TTS, STT, txt2img, txt2audio, img2img, txt2vid, vid2vid... well you get the idea and i can add new data anytime i want by uploading a sample to the interpreter. But i don't think my solution is optimal for most. My personal lab stack is 3 servers with xeon platinum x2 each with 3090ti x4 each at 1.5tb ram each and a 240tb nas (if you think thats sick i wish i could tell you about the production stacks in the colo)
 
  • Like
Reactions: felldude
F

felldude

Active Member
Aug 26, 2017
572
1,724
cryptonaut said:
As of today i have officially migrated completely away from fix point models, all my data point and tokenization is now dynamically stored in a graph db fully multimodal so now i have a unified API interface for TTS, STT, txt2img, txt2audio, img2img, txt2vid, vid2vid... well you get the idea and i can add new data anytime i want by uploading a sample to the interpreter. But i don't think my solution is optimal for most. My personal lab stack is 3 servers with xeon platinum x2 each with 3090ti x4 each at 1.5tb ram each and a 240tb nas (if you think thats sick i wish i could tell you about the production stacks in the colo)
Click to expand...
An 8 stack of A100 80GB if your planning on doing any real training :LOL:
That was the recommended setup back for Spade or Style-Gan training

Your setup sounds like it could actually do some work, your still in CUDA realm and not tensor with your setup though right?

And all this to do the Mandelbrot calculation 1e+18 times. Seriously is their any .cu files outside of mandelbrot....

Did you make your own UI?

Final edit (Hopefully): What is your average IT/per sec training a lora at 512x512 mine averages 2.5 over 18 minutes for most. (With one 3050 at 98% load)
 
Last edited:
N

NeoVlahos

Member
Jun 17, 2017
187
269
Just taking up some space again (@sharlotte "No it isn't but thank you for your kind words") anyway some more pics.
Sure thing! i will make the damn thing give me what i want eventually o_O:):giggle:

00018.png 00020.jpg grid-0008.png
You don't have permission to view the spoiler content. Log in or register now.
i am just a simple night guard, guarding those who really have access to true Ai my bosses :cry: they do through a big bone sometimes though :sneaky:
 
  • Like
Reactions: sharlotte, felldude and Whitekink
You must log in or register to reply here.
Top Bottom
4.80 star(s) 6 Votes