This year, the OpenAI company—the creators of ChatGPT—released Sora, the first AI software that can create videos. Some of the demonstrations they showed were incredibly realistic, ranging from a puppy playing in the snow to a view from an underground train that was navigating the city. It was impressive as well as terrifying, as I explained in my first thoughts here.
Since then, Sora News has been mostly quiet. The company hasn’t released the platform for testing by the public, so we’ve had a few first-hand experiences to draw from. However, that’s about to change this week. The OpenAI team is not only making OpenAI Sora more accessible, but they’re also opening its AI software to anybody who has an account on ChatGPT Plus accounts. The future is set to go wild.
What’s New in Sora
OpenAI announced the news on Monday after an earlier announcement from Marques Brownlee. If you’ve followed the original Sora announcement, this list of examples below will surprise you. Short story: Sora can be prompted to create photorealistic video clips in a variety of situations: OpenAI and Marques Brownlee both demonstrated drone footage of cliffs as well as animals in the wild, as well as people doing their work “on camera.” What’s new is a variety of new options OpenAI has incorporated into the Sora program overall.
Then there’s “Storyboard,” a sort of video editor that allows you to join multiple video prompts to create longer videos on a particular topic. For instance, you can create a prompt that asks for a crane to stand in water and a second soliciting that the crane drops its head in the water. In the end, Sora will combine those two prompts into a single video. “Recut” also acts as a video editor; however, in this case, you can select the specific portion of your video and then ask Sora to expand it. “Remix” opens a new prompt box, allowing you to request modifications in an already existing movie. (You can select your preferred “strength” of the remix also, which determines what percentage of your video is altered from the prompt.) Additionally, “Blend” lets you select to transform the theme of one video into another. Sora’s example is having the butterfly in one clip transform into an orchid in another video.
Of all the new features, the Storyboard one seems to be among the most fascinating. It seems to be an effective way to force the AI to create a complex scene that has multiple actions since trying to squeeze everything into one prompt is likely to be ineffective. Remix can also be useful to fine-tune the elements of a film without having to discard the original production. But at the end of it all, our attraction to the model stems from its fundamental purpose: you provide Sora with a concept for a video, and it produces the video for you. You can also upload a picture from your collection, and Sora creates animations of the inanimate objects into a moving image.
Sora in Action
If you send a request, the video will be included in your “queue” for processing. The length of time the video will take to create depends on the settings you choose, which include duration, resolution, and the number of variations you create, for example: I’m on an ordinary ChatGPT Plus subscription, and I’m limited to a resolution of 720p and a maximum duration of 5 seconds. ChatGPT Pro users can increase the resolution up to 1080p, make videos up to 20 seconds, and produce up to four different variations of their videos.
For me, it seems that everyone and their mom is trying to make use of Sora in the present. My initial and sole prompt attempt (“tracking video of a taxi moving through a city center”) was in limbo processing all the time I was writing this article. OpenAI has halted account creation at present since there are too many users trying to connect to their video creator.
My video eventually generated some footage, but it was rough. The quality of the video for the taxi, as well as the city, was quite high (again, extremely realistic); however, the movements of the taxi were across the board. The taxi first drove in reverse and then changed into a car in a different direction before speeding off as another taxi sped into the foreground. (My taxi was also a goner into thin air, but the new one did not possess a trunk; rather, it had two fronts.)
Because it takes so long to get Sora to produce videos, it’s beneficial to observe someone, for example, Marques Brownlee, who had a go at testing this software. In his evaluation of Sora, it is apparent that the tool is struggling to stay clear of the common problems with AI-generated videos. The videos may appear realistic, but they lack real-world realism when moving. Sora frequently confuses which leg is in the front and which needs to be back in walks and will also “forget” about objects altogether. When Brownlee wanted to see the video of a tech expert with a smartphone in their hands The reviewer had the two phones in his hands, and the other one vanished without explanation. Some elements of a video might run in slow motion, whereas others are running at normal speed, which appears odd to the untrained eye. These kinds of glitches are present in the majority of Sora outputs I’ve seen. If you’re searching at them, you’ll notice them and bring attention to the fakeness of the film.
This is the case with “low-quality” videos, too, for example, like the newest versions of CCTV or security camera footage. Vehicles collide and vanish, or people are moving in a way that isn’t realistic. However, the poor resolution of the videos makes it easy to create fakes. If Sora can understand the laws of physics, then people are going to spend the whole day creating CCTV footage that’s not there.
According to Brownlee’s observations, the things Sora currently excels at aren’t realistic at all. Motion graphics, for instance, usually look great, like some videos of animated characters and animations. A sketch animation of the Empire State Building looks like something from the Netflix series ‘ intro, for instance. Also, when Brownlee posted an image of leaf characters created by DALL-E, Sora animated the image in a manner that seemed to be believable. It’s more difficult to overlook the flaws in a video that doesn’t claim to be genuine in any way.
Sora is also proficient in creating drone and tracking images: a drone view taken of Mount Fuji or the Golden Gate Bridge appears natural and photorealistic. If you pay attention, you may notice a few imperfections and glitches, like waves that don’t behave exactly as they should. However, you can probably sneak these photos into shows or films without a lot (or all) people being aware of it.
What do we do from this Point?
Sora terrified me in February when it was first made public. Since the announcement, I’m still in awe; however, not because the quality of the videos is superior. Based on what I’ve seen today, the quality is similar, albeit with innovative AI features that you can apply to modify the videos. The realism remains in the right proportions, but so are the flaws, and there are numerous.
The thing that scares me is the accessibility issue: Once OpenAI is able to meet the demand, Sora will be available to everyone who has a ChatGPT Plus membership. For just $20, you get access to software that can create up to 50 videos of five seconds every month. Five seconds isn’t a lot of time, obviously, so without a little savvy, the videos aren’t likely to be the ones that cause the most damage.
This is where ChatGPT Pro comes in. The plan is costlier ($200 monthly), and, however, for $200, you’ll be able to create as many as 500 video clips. Each can be as high as 1080p and 20 seconds long. OpenAI states that you can download these videos without a watermark, making detection much more difficult.
Most of us don’t want to sign up for Pro for this reason; however, $200 isn’t enough to deter criminals who seek to spread false information. Imagine the next big conflict that is polarizing, fueled by an influx of video clips that “prove” what happened one way or the other, but they’re not authentic in any way. OpenAI has some security features built in, such as preventing copyrighted material or famous figures from appearing in a video; however, we’ll be able to see how these safeguards are working in the real world.
How to try Sora
As of now, the account creation feature isn’t currently available on Sora; however, this could be changing in the near future. If you’re looking to try Sora out, visit sora.com. Click here to log in, and then sign in with the details of your ChatGPT account. Be aware that you must have a ChatGPT Plus account ($20 per month) or a ChatGPT Pro account ($200 per month) to access Sora.
