AI startup it’s not Despite reports of disruption at OpenAI dominating the airwaves, OpenAI appears to be on track to stick to its product roadmap this week.
See also: Stability AI announced this afternoon announced Stable Video Diffusion is an AI model that animates existing images to generate videos. Based on Stability’s existing Stable Diffusion text-to-image model, Stable Video Diffusion is one of the few open source and even commercially available video generation models.
But not for everyone.
Stable Video Diffusion is currently in what Stability describes as a “research preview.” Those wishing to run the model may be interested in Stable Video Diffusion’s intended applications (e.g., “educational or creative tools,” “design or other artistic processes,” etc.) and unintended applications (e.g., “representation of people or events,” etc.). (a representation of fact or truth).
how Other such AI research previews – include Unique stability — Historically, I wouldn’t be surprised if this model started circulating on the dark web in a short period of time. If this were to happen, I would be concerned that Stable Video could be exploited, as it does not appear to have a built-in content filter. Once Stable Diffusion was released, it didn’t take long for actors with questionable intentions to use it to create non-consensual deepfake porn of themselves.
But I digress.
Stable Video Diffusion actually comes in two model formats: SVD and SVD-XT. The first SVD converts a still image into 14 frames of 576 × 1024 video. SVD-XT uses the same architecture but increases the number of frames to 24. Both can produce video at 3 to 30 frames per second.
according to white paper Released at the same time as Stable Video Diffusion, SVD and SVD-XT are first trained on a data set of millions of videos and then “fine-tuned” on a much smaller set of hundreds of thousands to about 1 million clips. it was done. It’s not immediately clear where these videos come from, and the paper suggests that many come from public research datasets, so determining if any are under copyright It’s impossible to do. If so, users of Stability and Stable Video Diffusion could be exposed to legal and ethical challenges over usage rights. Time will tell.
Image credits: Stability AI
Whatever the source of the training data, the models (both SVD and SVD-XT) produce fairly high-quality 4-second clips. My guess is that the selected samples on Stability’s blog perfectly match the output from Meta’s recent video generation model, as well as the AI ​​generation examples we’ve seen from Google and AI startups Runway and AI. There is likely to be. pika research institute.
However, there are limits to the spread of stable videos. Stability has been transparent about this, writing on the model’s “Hug Face” page — of page From where researchers can apply for access to stable video dissemination, models can generate video without movement or slow camera pans, control it with text, render text (at least not readable), It is not possible to consistently generate faces or people “properly.”
Although still in its early stages, Stability says the model is highly extensible and can be adapted to use cases such as generating 360-degree views of objects.
So how does Stable Video Diffusion evolve? Stability offers a variety of models that “build and extend” SVD and SVD-XT, and a “text-” model that “builds and extends” SVD and SVD-XT, as well as a “text- to-video” tool. The ultimate goal appears to be commercialization. Stability rightly points out that Stable Video Diffusion has potential applications in “advertising, education, entertainment, and more.”
Indeed, Stability is poised to be a hit as startup investors ramp up the pressure.
April, Semaphor report Stability AI has run out of cash and spurred an executive hunt to boost sales. According to Forbes, the company repeatedly delayed paying wages and payroll taxes or didn’t pay them at all, and his AWS, which Stability uses for calculations to train its models, gave up access to Stability’s GPU instances. He is threatening to cancel it.
![Popularization of stable video](https://i0.wp.com/techcrunch.com/wp-content/uploads/2023/11/output_tile.gif?resize=749%2C421&ssl=1)
Image credits: Stability AI
Recent stability AI raised The company raised $25 million through convertible debt (i.e., debt that converts into equity), bringing total funding to more than $125 million. However, it has not completed new financing at a higher valuation. The startup was last valued at $1 billion. Stability is said to quadruple in the coming months, even though revenues remain low and burn rates are high.
Recently, stability has taken a new hit. departure Ed Newton-Rex served as VP of audio at the startup for just over a year, and played a key role in launching Stable Audio, Stability’s music generation tool. In his open letter, Newton-Rex cited disagreements over copyright and how copyrighted data should and should not be used to train AI models. He said he retired from Stability.
Source: techcrunch.com