AI stability has improved by leaps and bounds again: amazing new video generation demonstration, unanimously recognized by netizens

Stable Diffusion is coming back?

Stability AI CEO Emad Mostaque’s latest tweets and four videos have attracted countless imaginations.

Many netizens suspect that this is a demo of the new version of Stable Video Diffusion.

Because from the effect point of view, the picture clarity, consistency and smoothness are all amazing.

For example, this stuffed bear goes diving with swimming goggles.

The stuffed bear moves very smoothly and even blinks. There are also rich details of the sea water behind it.

In addition, Emad Mostaque himself has disappeared for several weeks since January 1 this year and has never appeared on social networks.

#Netizens feel that this wave is more like holding back big news. Stability AI is back?

(I once thought that Stability AI had fallen behind in the competition)

Stable Diffusion is back?

Judging from the four-second demo released, it should be an upgrade to the SVD video generation function.

However, it is not clear whether it is based on text or image generation. Let’s feel the effect directly.

There is also this anime-style night street.

In November last year, Stable Video Diffusion (SVD) was officially launched. At that time, in addition to supporting images and text to video, it also supported 3D synthesis of objects.

According to the plan disclosed at the time, SVD was only used as a basic model, and it would continue to expand to establish the entire Stable Diffusion ecosystem.

At that time, netizens’ feedback on the Demo included incorrect lighting and shadow, and overall incoherence (video frames flickered between frames).

Looking at the current Demo, let’s compare it with a similar video: Green robot lying on the bed.

You can see that the details of the robot, the cups and lamps next to it, and even the folds of the bed, quilt, and pillows are richer.

And there is also obvious light and shadow mapping on the robot's face.

#However, more update information will have to wait until the official version is released.

It is worth mentioning that after this tweet, Emad Mostaque also posted a similar one.

No text, four still pictures.

Netizens expressed their expectations. "I hope this model can surpass SDXL in all aspects and better iterate in time." "Is this SD3?"

But What the hell is this earth tomato? ? ?

On the same day, their StableLM2 1.6B was released, a small language model with only 1.6 billion parameters. It is multilingual trained in English, Spanish, German, French, Portuguese and Dutch.

A few days ago, they just released Stable Code 3B, the first large language model since the new year.

With such frequent progress, is Stability AI really about to turn around?

Decisive Battle Video Generation

The reason why I want to talk about a turnaround is because Stability AI has indeed attracted much attention in the last wave of Wenshengtu track. The popularity of Stable Diffusion made it once the world's most popular AI unicorns to watch.

However, with the emergence of Midjourney and the frequent progress of OpenAI and Google, there are also internal unrest, including the CEO suspected of plagiarizing achievements, misappropriating public funds, the company's business model is unclear, and wages are in arrears. Stability has exploded. AI was once caught in a storm.

In November last year, it was officially announced that it would enter the video generation track. At that time, popular players such as runway and pika had already appeared.

Nowadays, a new version of SVD is suspected to be unveiled. The reason why it has attracted everyone's attention is that in addition to its own company, it is also closely related to the current high-energy video generation track.

Many netizens said bluntly: There are really waves of curls, one after another.

#Either it is iterative than the version release. During this period, new developments such as Alibaba, a major domestic manufacturer, have appeared frequently. Either from the generation effect, clarity, and smoothness. Some time ago, Moonvalley came out and shocked many people.

Or maybe focus on controllability, etc...

Isn’t this just a few days ago, Runway’s Gen-2 evolved again——

Just use 5 brushes and brush against a picture, and the birds that were originally stationary will start to move.

So in the video generation track, are you optimistic about Stability AI?

The above is the detailed content of AI stability has improved by leaps and bounds again: amazing new video generation demonstration, unanimously recognized by netizens. For more information, please follow other related articles on the PHP Chinese website!