Stable Diffusion 3 - An Amazing AI For Free!

📝 The paper "Scaling Rectified Flow Transformers for High-Resolution Image Synthesis" is available here:

📝 My paper on simulations that look almost like reality is available for free here:

Or this is the orig. Nature Physics link with clickable citations:

🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Alex Balfanz, Alex Haro, B Shang, Benji Rabhan, Bret Brizzee, Gaston Ingaramo, Gordon Child, Jace O'Brien, John Le, Kyle Davis, Lukas Biewald, Martin, Michael Albrecht, Michael Tedder, Owen Skarpness, Richard Putra Iskandar, Richard Sundvall, Taras Bobrovytsky, Ted Johnson, Thomas Krcmar, Tybie Fitzhugh, Ueli Gallizzi.

[KimForsberg]

I just love how "ok, this old technique from a few months ago" and then "here is the new revolutionary technique". The rate of evolution of machine learning is just insane.

[Robin_Goodfellow]

Stable Diffusion out here doing what OpenAI was supposed to do

[viddarkking]

We were already able to do extremely detailed and exact images with ControlNet and Loras.
I am absolutely salivating at what we'll be able to do once we can finetune SD3.

[NLite]

i signed up for the waitlist, can't wait to see what bullshittery I get to generate

[larion]

I just hope the new model is not too censored. Sometimes it feels like the more advanced any of these become, the less fun we're allowed to have with them. That has been the theme with big LLM's at least.

[TheSmurfboard]

As an SDXL user I’m giving this new version three thumbs up !

[paolettinic]

1:49 trypophobia triggered so badly I threw the papers

[CNC-Time-Lapse]

I can't wait for this one... Such a massive improvement over SD2 in so many ways. The community is going to make some crazy good models with these too.

[morgan]

SD is open source and have no any kind of censorship.

[KenDesigns]

I can already do so much in SD-1.5 with a mere 1B model, while SDXL is 3.5B, and SD-3 gets so much more capable with 8B. I know a LOT of work was put into the algorithm's optimizations, but AI seems like a field where size does matter. I can only imagine what SD would eventually be able to do with a 13B model. What a time to be alive! 😁

[EvanBoyar]

Interesting that the paper includes the avocado students in front of the chalkboard because that has an error in the text. There are seven "r"s on the chalkboard while the prompt only calls for six. I wonder if it still struggles with repeated letters.

[smorty]

WARNING for people with fear of holes: Skip the time between 1:48 and 2:00

[zdanee]

I still love how at every introduction your accent at your name completely changes back to Hungarian :) No AI voice generating model will ever be able to generate it quite like that!

[sinkler]

As a fan of the channel, I am making an NPC researcher in my D&D campaign, based on your charming personality. :)
Can you please share what laws 2 & 4 are? These quotes are priceless. 🙏

[encyclopath]

I hope there will be the option to use the legacy text generation when they finally perfect it. it’s so endearing.

[scooterboom]

I love youre videos man! hope you keep reminding me that the time I'm alive in is incredible!

[marcfruchtman]

Thanks for the review. Yea, Text in the image generation software has been "underwhelming"... to say the least. This is a huge improvement.

[latt.qcd]

This looks like a jump over 1.5 that will actually be worth the upgrade.

[yuryeuceda]

Last week i dream about a place like a castle and when I wake up I was thinking how the brain can create this kind of scenes hyper realistic and taken from nowhere. I was thinking if we can access the brain in such a way we can generate this scenes from our brain and put them on video it could be amazing creativity

[aderitosilvachannel]

I'm eager to see how SD3 deals with hands and teeth. That seems to have been the curse of AI generated images for a while.