DiT: The Secret Sauce of OpenAI's Sora & Stable Diffusion 3 

Don't miss out on these exciting upgrades designed to elevate your content creation experience with DomoAI! Go try out: discord.gg/sPEqFUTn7n

Are you telling me Stable diffusion had ADHD 💀

The hell with the Frieren thumbnail lmao

5:43, OpenAI itself said that the more compute they threw at sora, the better the results got, so you are right in that the compute here does absolutely matter!

Hey, great video as always. Would love to see a deep dive video into the Stable Diffusion 3 architecture and other DIT methods!

Those facebook users are probably bots, because we are reaching the dead internet theory

yeah it would be nice to hear more about DiT architectures

Just letting you know, Yes, I would like to see your video on DiffiT and HDiT architecture if you make one! Love your videos!!

In the interview, the developers mentioned that Sora will still need quite a while until its release. In the subsequent interview, however, the chief technology officer stated that there might be expanded access this year, possibly even in the coming months. I believe this is yet another instance of the conflict between scientific caution and the realization that, if high prices are initially charged as announced, there can be significant earnings.

The "Sigmoid Curve" is FAR too misleading a representation to actually give people any reliable indication of where we are, unless their goal is a skeptical narrative. It's impossible to tell where we are on the curve, or what the curve truly looks like at this time, as it's significantly different with every advance, and how we arbitrarily classify advances also changes the appearance of the curves. Such representations can ONLY be applied retrospectively, when analyzing the past, they have no value for reliable prediction. Maybe the curve has multiple slow spots along it, or maybe multiple sigmoid curves chain together. Somewhere I saw an explanation, that overlayed a whole bunch of sigmoid curves related to recent technological advances, and when you average out all their overlaps, you end up with the same exponential singularity curve, guys like Kurzweil predicted. Tho I can't seem to find that.

You reached the pick of chill in this video. I like the vibe.

I definitely would like for you to explore the other DiTs

Definitely want to see an indepth look at DiT

Frieren thumbnail, nice

+1 for the follow-up video on the more technical details of DiT.

I'd love to have a video from you on those architectures!

My exact reaction when started usong if ai in comfyui

The only vid someone needs to understand current ai situation

"Can you even tell which one is the real and fake image?" - Uh, yes, it was pretty obvious. I didn't even had to 'nitpick' about it, I could tell pretty much instantly. I completely disagree with the notion we are high up on the curve. If you actually work with AI, not just use MidJourney. I'm talking about Krita and text completion models (for example to make Agents'), you will see there is still much progress day over day. I do generate a couple thousand of images each and every day (work related). So I do have quite some experience from seeing so many of them. I haven't even seen an upscaler yet that satisfies my requirements. The only way to get better upscales right now is more VRAM on a single device. Which isn't practical. My current workflow is pretty convoluted. I have multiple GPU's all rendering different image layers, each with their own focus (like attention). I have to use multiple GPU's, one for each layer to make it viable. Rendering each layer one after another on a single GPU would not be practical. - So instead of using a GPU with tons of VRAM or use a very complex workflow with multiple GPU's, I'm still seeing great progress being made in efficiency. for example SD Cascade. I think your focus on what progress is, is too narrow. Don't just look at what is possible, you have to take in to account how much progress there is being made in what is possible with limited resources. Which isn't just the amount of GPU's either. Also the amount of, and quality of, the required input.

Love how the ai took over even sponsors

Domo AI can get expensive very fast. Went through a month's tokens from the basic subscription in an hour transforming my client's 30 sec ad into an anime. The results were good.

I've been waiting for this video!

THE THUMBNAIL THO LOL i lov it

Did you see the MIT paper that just published? Using "Distribution Matching Distillation" (DMD), 30x faster image generation is achieved vs. Stable Diffusion, and at the same/higher quality of image. How's that for near the top of sigmoid curve 😉

The only thing they added was space time relation is such an understatement.

Venturing into the world of storytelling and creative videos, VideoGPT becomes the invisible hand that refines my content, making it resonate with a professional vibe.

Ability to generate more complex scenes is cool but I don't see much difference between SD3 and SDXL in terms of image quality.

I can still tell the difference between fake and real images but that's because I study photos and paint them. Something about the noise and the lighting is really off in the first image, most people wouldn't light something slightly green since it's not a natural type of light. I think if you don't have a background with visual art or something similar, it's probably really difficult to tell real and fake images apart. (0:14)

Brilliant style. All the best insh'Allah :)

To me, the biggest hurdles to overcome with image generation are character consistency (which is exactly what it sounds like) and object transfer. (The ability to select a specific object in one image, such as a shirt, and have it be included as part of the resultant generation.) AI image generation doesn’t need to look perfect, and I’m not to sure how much of a return companies will get over marginally increasing quality, when solving the issues I described above would lead to greater direct application. One immediate possibility would be how useful these advances would be for creating ai generated comics or animations. If I were a betting man, I’d say these two concepts are what most AI image generation companies are working on right now.

Just watched your video on Mamba. After watching this it makes me wonder about the Mamba model being used more and more for it's precision. Things like pixel art or voxel style need a lot more precision than regular diffusion or other image/video generation has. At least with what I've seen. Although part of that is that how people make that is by giving themselves limiters such as only drawing on a bitmap within certain parameters or with programs that automates those parameters. That won't be as much of a limiting factor if AI is now getting access to a command line which means it's only a matter of time before the AI has access to a program where the parameters it works with are constrained and narrowed to get the results people want.

I am starting to get confused between these channels

I wish you would go into more detail on the actual mechanisms for those who can follow it.

It's much harder to tell what's fireship and what's bycloud on recommended feed these days than AI image progress :D

yes.... dive into DiT, Diff it and HDit, CorrDiff, eDiff-I etc... 😊

Banger

lower pic was Pulp Fiction so the other one had to be fake

Can you explain the DiT usage in Pixart sigma?

background music?

Transformers is a very good movie. You can't go wrong to watch it with someone (except for some movie snobs). Transformers and chill is the new 10/10.

I'd like to know more about diffIT and hdit ^ _ ^

Am I crazy, or is the editing oddly similar to Fireship?

DiffiT video WHEN?

this one is a hot one

pulp fiction reference nice

Please do DiffIT and HDiT videos!

If you like DomoAI, you can find the open source communities that they take all their workflows from, and learn to do it yourself for free :)

Fireship but 8 times longer

The only thing you could really say is we are at the top of the 'current' sigmoid curve, until it's broken by yet another sigmoid curve. Using sigmoid curves doesn't work for predicting the future, they're only useful for looking at past processes.

how about MoE DiT SSM b1.58 transformers

What is that game in @5:07?

Is this fireship second channel

"attention is that we need" i see what you did there 😅

Just because companies can afford lots of compute does not mean that they have a commercially viable/environmentally sustainable product. These are great "experiments" for sure and kudos to the amazing work being done by these scientists and engineers. But to me, these are still experimental. I would call an AI product successful when I can run it on my phone's compute or something like an OS that can greatly augment general purpose tasks without ever having to connect to a cloud subscription.

Dive into DiffiT HDiT plz 🙏🏼

Finally

lmao I livestreamed teaching the chatbots to use stable diffusion XD

GPT-6 -> AGI - Supercomputers at consumer level → Quantum Computing -> ASI

why do you only have on screen subtitles in certain parts of your videos? it kinda defeats the purpose of subtitles in my opinion.

Man i picked the One below and got stumped whent he above one was fake. I swead these are getting so good that someday you can do it with both images as real and I would still not get it.

Can I still participate in the giveaway coz I need a PC and all I have is a i5 7thGen laptop

Stability AI must be saved.

I predict some r34 content

I love how some millionaire guy was like let's just throw money and it and it actually works

its finally over for artists

Ringo and Honey Bunny are the real image. Not shown: wallet with "bad mother fucker" stamped on it.

The most probable cause is that we are limiting the growth of ai, at least we are starving it with novelty

The thumbnail......😂

I actually helped figure this out :3

I'm worried SD3 will never release though.

Why does this video type looks like fireship's video.... Did firship stole the idea(or inspired) from this channel... Or this channel stole it(or inspired) from fireship????

I think we are far from perfecting image generation even with SD3, it still struggles with background details and hands to a lesser extent

@4:06 - the other ones don't follow your prompt because they don't want you to compete with them - the others are heavily funded by corporations - all the wealthy are aligned and they don't like competition...

Oh boy I can’t wait for easy to make AI blackmail material and mass unemployment!

Their secret, I think, is that they are open source... because PEOPLE THEMSELVES MODIFY THIS AND IMPROVE IT AS THEY WANT.

i dont think its the sigmoid curve of ai image generation its the sigmoid curve of agi

I'm worried a 4090 is not gonna be enough for stable diffusion 3 anymore lol. And knowing nvidia they probably won't be releasing a GPU with more than 24GB of ram any time soon

It's all fun n games until you realize that a Sora 2 min 720 p blurry and unintentional artifact filled video requires 720k H100 GPU's @ $30k each whilst Avengers Endgame using Autodesk Maya generates 4k masterpieces with GPU's that you can buy for $80.00 on eBay and 3d animators that work for less than 3 H100's. And Avengers Endgame generated 2.8 BILLION dollars from a $356 million production cost and paid a lot of human's grocery bills.

WHITE THEME DISCORD WTF

AI is genuinely getting to the point where it isn't even fun anymore. the only reason I ever found it interesting was that I thought it wouldn't really go anywhere, but now I see it so often, and I'm constantly trying to figure out if something is "AI". Like I just want it to stop now before it starts taking jobs from people. Especially considering that, basically every field I'm talented in is being attacked by these corporations looking to give even less money to the ones who deserve it.

TBH I'm dumbfounded how much you rely on low res gifs with text on top while talking about image/video generation. It seems that it is basically useless to you or you have no idea how to pivot it to your workflow.

ai art devs should focus on composition not quality. at this point the only composition method is control net which works great with humans but...what about everything else buildings, landscapes, dragons, eldritch creatures.

Why does He talk like his eating a hot potato

Stolen Fireship thumbnails to farm views! Clever little kid!

Makes a whole video about 'DiT', never actually defines the term... 🙄

Humans will just get more lazier, let's be real. It's like how the anime industry pushed towards CGI use, in a way.

Stealing other's thumbnail styles is bad