Stable Diffusion 3 Stunning new Images - Sora delayed - AI news

#### Links from the Video ####
SD3 Images:
Myth Maker Ai Image:
Sora Delayed:
SDXL Lightning + Manual Pose
Ai 3D Project:

#### Join and Support me ####

[mattc]

SORA = We used 1000 GPU’s in our cloud to show what video AI can be created, but there isn’t enough GPU’s on the planet for us to release this even under a $3000 a month subscription. See you in 3 years folks is my guess on Sora release.

[onufoms]

SD3 needs control and consistency. You can't say something is the perfect final evolution of image generation when it only gives you an output without obvious imperfections 1/100 times.

[MarcoEscobedo]

I think that solving the definitive problem with the hands will be a significant advance incorporated into this Sd3 that I would like to see in the future.

[diomymio]

1:54 - oh, my, Drizzt portrait! love forever! ❤Drizzt and Guenhwyvar

[TheReferrer]

Drizzt Do'Urden and Guenhwyvar used to love those books when I was a kid.

[NinetooNine]

I am pretty sure they showed SORA way sooner than they intended to. If you remember they did it the same day Google showed off Gemini 1.5. Open AI seems to have a habit of news-bombing other AI companies' announcements by showing something they are working on.

[I-Dophler]

Deine Einblicke sind wirklich inspirierend! Danke für die aufschlussreiche Präsentation der Zukunft von KI. Freue mich schon auf weitere Updates!

Your insights are truly inspiring! Thank you for the insightful presentation of the future of AI. I'm already looking forward to further updates!

[KrakenCMT]

Emad isn't thinking like an artist. An artist looks at their work and always sees opportunity for improvement. :) I'm sure that either Emad or the community will continue to keep SD improving for years to come.

[Deadgray]

We will see when SD3 will be in our hands, for now IMO those are all cherry picked images. Can't compare consistency without generating 10 images with different seeds.
Also workflows made on open source software, with nodes shared with you for free... behind paywall is big nope.

[realmike]

@OlivoSarikas, you should read the book "Homeland" by R.A. Salvatore. That was not just "some dark elf" on that picture. You can thank me later.

[ChrlzMaraz]

They didn't fix the hands, give more control over perspective, be able to better separate characters and subjects, etc... etc... We are HARDLY done.

[petitemasque]

Love the Drizzt portrait, I hope you don't need Skynet to run it and train it.

[AIchemywithXerophayze-jtgg]

I can't wait for this model to come out so I can try the prompts I generate from my prompt generator on it. Right now they work awesome with SDXL, 1.5 and 2.1. it'll be interesting to see just how much better SD3 is.

[danielchoritz]

true, you will always need better understanding and creative power for image creation-cool would be to describe your wanted image and then choose between 4 variants until it hits your need and can put out the stuff with changing viewpoints, scene etc.

[kernsanders]

Spoke to Stability employee, they said it is gonna be released THIS year, they just cant say when. Which is why they avoiding to say it's "close". As they cant say for sure if its days, weeks, months. Do note they are censoring the models so its purely SFW. So one of the reasons even though they have models, they want to "sanitize" it before release. Emad announced on twitter that the SD3 Turbo 8b parameter preview version is done cooking and they might be looking for preview testers. Also note SD3 at release will be a democratized models, so its from 800Mil to 8Bil parameters versions, so that any GPU can run it. So it will have models larger than SDXL and models smaller than SD1.5. Of course the accuracy coherent previews are done on the 8B models.

But 100% agree with you, as soon as we have models that I can perfectly describe 5 different characters and have them interacting with each other without them becoming mutant collective versions of each other, this won't be great for a Final No More Big Version release. Unfortunately if they are gonna go with that, it means this will be last big release for public access models... which is horrible imo. So the big corpos will keep improving their models and public accessed locally run models will be left in the dust. Such is life I guess... Hope one day someone can come in and pick up where SD left off. After Mistral buyout, I kinda was expecting this news. Seems the big corpos are working overtime and spending billions to take away private accessed AI technology. Microsoft, Apple, etc they all seem to be strategizing together behind the scenes and buying up all open source stuff and closing it off. Not sure what Stability's deal is with them stopping to make open source access models. It was their mission to bring AI to the people. At least they did in fact do that before closing shop. Still... sad to see...

[gnoel]

Everyone is racing to get the graphics of RDR2 or Cyberpunk but what we need is the control, precision and polish of Hades or Counter-Strike.

We can already easily achieve crazy image quality with upscalers, the biggest drawback of AI is the pain of detailed control of the image and consistency of character(s). Only then will we be able to say we mastered Image AI.

[stuartclements]

That last showcase of the future of AI is unbelievable. I'm excited to see more, and thank you for keeping us all updated!

[jepperisum]

Funny that none of my favorite AI'ers beside Scott mentions anything about that SD3 comes without unet. Really curious to know more about the implications of this. (Meant as a loving reminder. Keep up your awesome work)

[shantyguy]

Not just the US but over 45% of the worlds population will be voting in their respective countries this year. That has been a major talking point for the InfoSec community this year. Simply a guess, but maybe SORA's delay has something to do with this.

[CoconutPete]

anyone know if SD3 will be able to be run on A1111? or is that a dumb question?