Stable Diffusion 3: Scaling Rectified Flow Transformers for High-Resolution Image Synthesis

  Рет қаралды 2,266

Gabriel Mongaras

Gabriel Mongaras

Күн бұрын

Website paper: stability.ai/news/stable-diff...
Paper: arxiv.org/abs/2403.03206
My notes: drive.google.com/file/d/1n8rS...
00:00 Intro
01:58 DDPM
13:16 ODE/SDE formulation and score
18:09 ODE intuition
21:38 Rectified Flows
27:46 Sampling from a diffusion model
29:16 Going to the latent space
32:17 CLIP
37:53 Model architecture
56:18 Results and stuff

Пікірлер: 5
@TTTrouble
@TTTrouble 3 ай бұрын
Man I took a break from getting into the weeds of the AI papers but I really appreciate that you’re still at it man, and it inspires me to jump back into the jungle. You’ve definitely been a fantastic source of knowledge and helped me break down some of this stuff in a really meaningful way. Keep up the great work!
@kevinxu9562
@kevinxu9562 3 ай бұрын
GOATED damn thank you so much for making a video on this! Timing is goated, just started going through your diffusion series as I'm trying to build a diffusion model!
@mathiasbang1999
@mathiasbang1999 2 ай бұрын
Hey I was wondering if you could clarify something for me. You say that the [154; 4096] matrix holds the "fine grained" information, but when explaining the MM-DiT block setup Y is marked as fine grained information. It does seem to make more sense for the Y to be fine grained information in my opinion as it is post reduction information, however as I am not entirely sure I would love for you to maybe correct me on that :). Really appreciate the video! makes a lot of sense overall
@alexalex-lz8sg
@alexalex-lz8sg 2 ай бұрын
Cool, what about latent adversarial diffusion distillation(LADD) video?
@gabrielmongaras
@gabrielmongaras 2 ай бұрын
Oh yea that was a good paper. Lemmie maybe a video on that. This week seemed a bit lacking in terms of papers :/
OpenAI Sora and DiTs: Scalable Diffusion Models with Transformers
1:02:38
Stable Diffusion explained (in less than 10 minutes)
9:56
Render Realm
Рет қаралды 3,9 М.
Homemade Professional Spy Trick To Unlock A Phone 🔍
00:55
Crafty Champions
Рет қаралды 58 МЛН
Tom & Jerry !! 😂😂
00:59
Tibo InShape
Рет қаралды 55 МЛН
I wish I could change THIS fast! 🤣
00:33
America's Got Talent
Рет қаралды 75 МЛН
How I Understand Flow Matching
16:25
Jia-Bin Huang
Рет қаралды 3,7 М.
Stable Diffusion - How to build amazing images with AI
44:59
Serrano.Academy
Рет қаралды 16 М.
Flow Matching for Generative Modeling (Paper Explained)
56:16
Yannic Kilcher
Рет қаралды 39 М.
How I Understand Diffusion Models
17:39
Jia-Bin Huang
Рет қаралды 22 М.