Stable Diffusion 3 Takes On Midjourney & DALL-E 3

Stability AI has done it again and is showing off a preview of Stable Diffusion 3.

SD3 is an entirely new model built on top of a new transformer architecture. This provides higher quality, faster training, and most of all, top-notch prompt following and adherence.

Today we take a look at the stable diffusion 3 examples and see how they stack up to SDXL, Dall-E 3, Midjourney V6, and even Stable Cascade.

[hleet]

I hope SD3 will be as fast as stable cascade (maybe more) while keeping old low vram card compatiblity.
It would be so cool if there was a kind of conversion software to extract or transform SD1.5 Lora into SD3 !
It's so a waste when SDXL didn't accept SD1.5 Loras ...

[onfire]

This is exciting progress! I agree with all the comments you made at the end. Google did drop the ball. People i work with are looking into recommending that clients integrate with google Ai systems on their sites. I told them they should hold up on that for now. I was ignored. People want to be the coolest out there so this stuff gets pushed when its not ready or viable. Google will eventually have the most powerful system in my mind to develop Ai and they will crush in the future. But what will be the advantages when they lock you into an ideological hole with the way they develop around it? Im looking forward to seeing where they go with it but wont hold my breath on it being open and/or non-ideological ever.

[polystormstudio]

Count me in as a skeptic until proven otherwise. SDXL was a letdown, and 2.1 didn't hit the mark. These pre-release teasers feel more like hype machines than anything else, especially with the constant barrage of 'revolutionary' AI updates that are showcased with cherry picked images. At this point, I'll reserve my excitement for when I can actually see and use version 3. Let's hope it lives up to the buzz and isn't just another drop in the ocean of AI promises.

[paulct]

12:20 - That part's only relevant if people choose to support it over comparatively easier to run models like SD1.5, SDXL Turbo. From what seems to be the case these models are only getting harder and harder to run and train for.

[angel_cheon-sa]

Can someone tell me what happened with google? i havent seen the news

[dev-jonathan]

Cool video! Guys, does anyone know of a model, checkpoint, or specific library for generative AI image models, suitable for UI/UX, web design, and user interface prototyping? The models I've found are very expensive in terms of ultra-realistic images, beautiful illustrations, etc. If anyone has an idea, it would be a huge help, it's for my college work

[jopansmark]

It's over for Midjourney and OpenAI.

[zapfska]

When are companies gonna learn not to hire these weird DEI ppl they ruin every company i know of look at bud light lol

[andrewcampbell]

It doesn't change everything. It changes something. Just so you know, the 'this changes everything..' title is getting really old. You have good content please don't spoil it with stupid titles and mouth wide open in shock thumbnails. You have to move on. This just doesn't work anymore.