每月参与
每月我在内容中的一个定期席位。订阅者进入队列——或当订阅者多于席位时随机抽签——并出现在视频本身、片尾字幕或描述中。
每月
价格面议
下一段
But how do AI images and videos actually work? | Guest video by Welch Labs 38 分钟 
We Took Our Food Delivery Man on His First Vacation 46 分钟 
The Gaming Market is Starting to Break. 23 分钟 
Jony Ive Shows The Real Process Behind the Ferrari Luce 46 分钟 
Why avalanches are most deadly when they stop - Simon Trautman 6 分钟 
What If I Fell Into A Black Hole? 20 分钟 
Digging A SECRET GARAGE Part 3 21 分钟 
Mac vs Windows - It's not close in 2026! 22 分钟 
An Engineer's Perspective on the Texas Floods 24 分钟 
All the Best Memories Are Hers 4 分钟 
The Wild Story of the Teton Dam Failure 20 分钟 ![Can we cross Manchester city centre using its forgotten Victorian tunnels? [PART 2]](https://adm.6-9.space/storage/films/posters/yt_can-we-cross-manchester-city-centre-using-its-forgotten-victorian-tunnels-part-2_1789704303.jpg)
Can we cross Manchester city centre using its forgotten Victorian tunnels? [PART 2] 40 分钟 
We Turned Our House Into A 5 Star Restaurant - CHUNKZ EDITION (FINALE) 46 分钟 
Is there a Priest Hole in Your House? | Fascinating Horror Shorts 1 分钟 
The Napoleonic Wars - OverSimplified (Part 1) 30 分钟 
Law professor calls the new pardon system “anarchy” #shorts 1 分钟 
The Weirdest Tool in Underwater Construction 18 分钟 
Why American Houses Are So Flimsy 15 分钟 
Stopping the Unstoppable 16 分钟 
The Most Insane Megaproject You Never Heard About 14 分钟 
Why Roads Get Washboards 18 分钟 
So You Want to Build a Tunnel... 21 分钟 
Every US Electrical Outlet Explained 22 分钟 
Name That Infrastructure (Ep. 3) #shorts 1 分钟 
What's Under Your Feet in New York City? 23 分钟 
From Garden to Jar | Preserving Cucumbers & Tomatoes for Winter 🌿🍅 26 分钟 
Man Builds Amazing WORKSHOP With Garden In His Backyard | Start To Finish By @machikane_base 
The price of proving you’re an adult 1 分钟 
The Real Reason We Should Revive Extinct Animals 16 分钟 
The Craziest Experiment Humans Have Ever Built 18 分钟 
The illusion of the ‘self’ | The Gray Area 43 分钟 
Hasan Piker on what Democrats get wrong | America, Actually 34 分钟 
Why This Robot Kills Weeds With Lasers 15 分钟 
Thomas, a tank engine, is not Thomas the Tank Engine. 18 分钟 
How Earthquake Bearings Work 19 分钟 
Edgar Allan Poe's story of the Red Death - Iseult Gillespie 8 分钟 
How Did Tuberculosis Get So Bad? #science #scishow #tuberculosis #health 1 分钟 
Are birds getting smaller? #shorts #science #scishow #climatechange 1 分钟 
Name That Infrastructure (Ep. 5) #shorts 1 分钟 
The Surprisingly Large Carbon Footprint of Rice #science #scishow #rice #climatechange 1 分钟 
How To Survive Being Eaten 10 分钟 
The Biggest Lie About AI 6 分钟 
Can Baldness Be Reversed? 9 分钟 
Anyone Can Make Amazing Games Now (Easy) 29 分钟 
Which of these is the best idea? 🤷♂️ #batmobile #batman #engineering 1 分钟 
Everything you need to know about bird flu - Benjamin Anderson 7 分钟 
How 3 enemies became one country 35 分钟 
How billionaires buy American elections 45 分钟 
UV printing on a Smith Blade? @Morpho.Global #uvprinting #engineering 2 分钟 
You SUCK at Prompting AI (Here's the secret) 24 分钟 
What is sumo, and why is it so popular? - Lee Thompson 6 分钟 But how do AI images and videos actually work? | Guest video by Welch Labs
2025 38 分钟
Diffusion models, CLIP, and the math of turning text into images Welch Labs Book: https://www.welchlabs.com/resources/imaginary-numbers-book Sections 0:00 - Intro 3:37 - CLIP 6:25 - Shared Embedding Space 8:16 - Diffusion Models & DDPM 11:44 - Learning Vector Fields 22:00 - DDIM 25:25 - Dall E 2 26:37 - Conditioning 30:02 - Guidance 33:39 - Negative Prompts 34:27 - Outro 35:32 - About guest videos Special Thanks to: Jonathan Ho - Jonathan is the Author of the DDPM paper and the Classifier Free Guidance Paper. https://arxiv.org/pdf/2006.11239 https://arxiv.org/pdf/2207.12598 Preetum Nakkiran - Preetum has an excellent introductory diffusion tutorial: https://arxiv.org/pdf/2406.08929 Chenyang Yuan - Many of the animations in this video were implemented using manim and Chenyang’s smalldiffusion library: https://github.com/yuanchenyang/smalldiffusion Cheyang also has a terrific tutorial and MIT course on diffusion models https://www.chenyang.co/diffusion.html https://www.practical-diffusion.org/ Other References All of Sander Dieleman’s diffusion blog posts are fantastic: https://sander.ai/ CLIP Paper: https://arxiv.org/pdf/2103.00020 DDIM Paper: https://arxiv.org/pdf/2010.02502 Score-Based Generative Modeling: https://arxiv.org/pdf/2011.13456 Wan2.1: https://github.com/Wan-Video/Wan2.1 Stable Diffusion: https://huggingface.co/stabilityai/stable-diffusion-2 Midjourney: https://www.midjourney.com/ Veo: https://deepmind.google/models/veo/ DallE 2 paper: https://cdn.openai.com/papers/dall-e-2.pdf Code for this video: https://github.com/stephencwelch/manim_videos/tree/master/_2025/sora Written by: Stephen Welch, with very helpful feedback from Grant Sanderson Produced by: Stephen Welch, Sam Baskin, and Pranav Gundu Technical Notes The noise videos in the opening have been passed through a VAE (actually, diffusion process happens in a compressed “latent” space), which acts very much like a video compressor - this is why the noise videos don’t look like pure salt and pepper. 6:15 CLIP: Although directly minimizing cosine similarity would push our vectors 180 degrees apart on a single batch, overall in practice, we need CLIP to maximize the uniformity of concepts over the hypersphere it's operating on. For this reason, we animated these vectors as orthogonal-ish. See: https://proceedings.mlr.press/v119/wang20k/wang20k.pdf Per Chenyang Yuan: at 10:15, the blurry image that results when removing random noise in DDPM is probably due to a mismatch in noise levels when calling the denoiser. When the denoiser is called on x_{t-1} during DDPM sampling, it is expected to have a certain noise level (let's call it sigma_{t-1}). If you generate x_{t-1} from x_t without adding noise, then the noise present in x_{t-1} is always smaller than sigma_{t-1}. This causes the denoiser to remove too much noise, thus pointing towards the mean of the dataset. The text conditioning input to stable diffusion is not the 512-dim text embedding vector, but the output of the layer before that, [with dimension 77x512](https://stackoverflow.com/a/79243065) For the vectors at 31:40 - Some implementations use f(x, t, cat) + alpha(f(x, t, cat) - f(x, t)), and some that do f(x, t) + alpha(f(x, t, cat) - f(x, t)), where an alpha value of 1 corresponds to no guidance. I chose the second format here to keep things simpler. At 30:30, the unconditional t=1 vector field looks a bit different from what it did at the 17:15 mark. This is the result of different models trained for different parts of the video, and likely a result of different random initializations. Premium Beat Music ID: EEDYZ3FP44YX8OWT
与此创作者合作
此视频背后创作者发布的套餐。
此创作者尚未设置费率。发送请求，我们将转达——在收取任何费用之前，价格会先经双方同意。