Original stable diffusion used frozen clip so no text encoder training
Original stable diffusion used frozen clip so no text encoder training
FJust training of unet
Cok
C
FNo pick unet
FMake step ratio 0
C
FStep ratio is text encoder training
FYes I think this
CThis would be fantastic
CI will run a test
FThere is also freeze but it is a new addition. I don't know how it affects
FNot in original papers
CWill try this first
HWow, this convo is getting very technical.. but im excited to try
CHomework before Dr.Fs video:
https://jalammar.github.io/illustrated-stable-diffusion/
Your a technical guy, you will survive
https://jalammar.github.io/illustrated-stable-diffusion/
Your a technical guy, you will survive
Translations: Vietnamese.
(V2 Nov 2022: Updated images for more precise description of forward diffusion. A few more images in this version)
AI image generation is the most recent AI capability blowing people’s minds (mine included). The ability to create striking visuals from text descriptions has a magical quality to it and points clearly t...
(V2 Nov 2022: Updated images for more precise description of forward diffusion. A few more images in this version)
AI image generation is the most recent AI capability blowing people’s minds (mine included). The ability to create striking visuals from text descriptions has a magical quality to it and points clearly t...
CYouTubeJay Alammar
Learn how AI image generation works. This video goes over the AI components of AI image generation models like Stable Diffusion and explains how they work and how they're trained.
Blog post: https://jalammar.github.io/illustrated-stable-diffusion/
---
Twitter: https://twitter.com/JayAlammar
Blog: https://jalammar.github.io/
Mailing List: htt...
Blog post: https://jalammar.github.io/illustrated-stable-diffusion/
---
Twitter: https://twitter.com/JayAlammar
Blog: https://jalammar.github.io/
Mailing List: htt...
CKnowledge is power, stay blessed, king!
Cmess with the best or die like the rest I say
HLol yes sir
Hthis is good, thanks for sharing
CI did 100 epoch, 200 steps
C
Cooo server got a boost
Hyeah, i boosted my favorite servers just now
C@Furkan Gözükara SECourses I think it worked
C
Fi watched this. i am planing to explain even with more depths. he skipped some parts :d
Foh 2 images?
C20
Fwow awesome ty so much
Rreddit
124 votes and 25 comments so far on Reddit
Flooks like now man is your subject :d
Fi saw this
Rhey yall, i saw this post about something that says stable diffusion but faster?
CI am trying
Rmy current wait times are like 2 minutes per image and saw that post and wondered if that would help improve my wait times possibly?
CI think best model of Zack so far
F
Ri don't really understand what that means but so in simple terms, that post isn't going to help me?
CSo what do we conclude from this?
Fi think there wont be like 4x speed
Fyou aint gonna get such speed unfortunately
Fhowever you can overclock
Fgpu
F10%
Rtheir discord server showcases it and its like insanely fast but i was just wondering if i could get even 1% of that speed
Rgenerations barely take a second from what i have tried intheir demo in their discord
C@Furkan Gözükara SECourses best method so far I think
HAnd you just left a basic filewords like in the video?
