Realistic-looking videos without sound are created by a new generative artificial intelligence (AI) from OpenAI. The service is called Sora and it creates videos of up to 20 seconds on demand. Sora is not currently available in German-speaking countries. But in North America, existing ChatGPT customers can enter instructions entirely as text or as rich text from static images or video. Video resolution is up to 1080p and can be ordered in square, portrait or landscape format.
Advertisement
Along with Sora, OpenAI has also released a younger sister model called Sora Turbo. It calculates faster, but doesn’t produce as good videos. ChatGPIT Plus subscribers can create up to 50 low-resolution (480p) or medium-resolution (720p) videos per month. Anyone who has a $200 ChatGPT Pro subscription can order ten times more videos in higher resolution and longer duration. Early next year, OpenAI would like to bring additional tariff models to the market to address different user groups.
The videos produced fluctuate between frightening and impressive realism and obvious errors. OpenAI openly admits that Sora isn’t perfect: “The Sora version we released has a number of limitations. It often creates unrealistic physics and struggles with long, complex processes,” the company writes. his announcementAll videos are watermarked by C2PA (Alliance for Content Provenance and Authenticity), and an OpenAI animation can also be seen at the bottom right.
there are rules
Uploading of recordings of minors is not allowed. Photographs of other people may only be used as part of the sign if the people depicted have consented. Additionally, Users must assure that they waive all necessary intellectual property rights. Depictions of violence and “explicit themes”, i.e. material that is not suitable for young people, are prohibited. OpenAI warned that anyone who uses Sora contrary to the rules may be blocked temporarily or permanently without refund of fees already paid.
user interface on sora website Shows videos that other users have ordered. The exact signals that reach the corresponding video can be called. OpenAI provides its users with insight into the art of AI signaling.
By the way, Sora video cannot be ordered only with individual signals; The interface allows you to put together a storyboard. A series of instructions can be entered to create a continuous video with different scenes.
mischief happens
Heize Online hasn’t been able to try out Sora yet, but American YouTuber Marques Brownlee was able to spend a week with it. In his examples it is not always clear whether a scene is actually filmed or Sora-generated. However, Brownlee also shows examples where errors cannot be ignored.
The text display is sometimes a mixture of characters, animals or objects, sometimes disappearing suddenly and without inspiration. Animals and objects also tend to move around each other. Sora also struggles with drawing the legs and general body positions of moving animals. With abstract representations and cartoon-style films, these shortcomings are less significant than with videos that seek to be realistic.
From Brownlee’s video we know that OpenAI offers downloads without watermarks if desired. This probably just refers to the visible animation, but we haven’t been able to determine that yet. The release of Sora is part of OpenAI’s “ShipMass”, twelve days of new innovation.
(ds)
