Sora’s AI video revolution is still a ways off

The first version of OpenAI’s Sora can produce footage of just about anything you throw at it—superheroes, cityscapes, animated puppies. This is an impressive first step for an artificial intelligence video generator. But the actual results are far from satisfactory, and many of the films are so riddled with oddities and inconsistencies that it’s hard to imagine anyone finding them of much use.

sola Published on Monday back Trailers from the past year foreshadows its capabilities. However, there are some hurdles before using the movie generation feature. One, Account creation is closed Within hours of release due to high demand. Those who successfully sign up will find that its features also require a subscription to unlock: a $20-per-month “Plus” membership lets you generate videos in 480p or 720p, with a length cap of 5 or 10 seconds, depending on the resolution . To unlock all content, including 1080p quality and 20-second videos, you’ll need to pay $200 per month for a “Pro” Sora subscription.

My Plus tier test results were less than impressive. Simple prompts with a limited description seem to work best—for example, “A cat plays with a ball of yarn,” produces a very realistic-looking cat bouncing excitedly across the floor. But Sora gave the cat a second tail for a while, and the yarn itself was so tight that it looked like poorly inserted CGI.

These visual problems are more frequent and apparent for complex cues that provide detailed scene descriptions. Making human movements look very natural is difficult: when I asked it to show me someone putting on makeup, hands were waving everywhere, and videos of people eating salads and sausage rolls were nightmarishly reminiscent. Viral AI clip of Will Smith inhaling spaghetti.

Sora includes an interesting storyboard feature that should help provide cue instructions for longer films. It’s similar to a movie editing timeline, allowing users to explain what they want Sora to produce every two seconds instead of inserting tons of description for the entire movie. It’s simple to use, but the results are worse. The more details I added, the more distortions and weirdness emerged.

There were a few things that really impressed me, though. Movies are produced faster than expected, and even clips that are 10 seconds long usually don’t last longer than 30 seconds. The patterns on the fur and textiles remain consistent even amid the fast-paced movement, and the lighting, shadow, and mirror effects generated by Sora do an excellent job of simulating the real thing. Sunlight shining through the windows will shine brilliantly and pass beautifully through all the materials you would expect. Even at low resolution, most objects have a high level of detail and don’t clutter into a pixelated mess.

Despite all its shortcomings, Sora performs better than runway artificial intelligencewhich is considered one of the better AI video generators for simulating realism. When the same prompt was entered on both platforms, Sora’s results looked more realistic and had far less visual distortion. Sora’s output quality is also on par with the demo I saw in October. Adobe’s Firefly video model In Adobe Max, although OpenAI clearly lacks the advantage of promising to produce output that is commercially safe. Adobe achieves this goal only by training its AI models on licensed or public domain content, and OpenAI does not follow this spirit.

[The above video was generated using Runway.AI using the same prompt I gave Sora.]

Nothing Sora generates from scratch is actually availablealthough. It’s definitely not suitable for entertainment or commercial work that requires narrative coherence, and you really have to use it in place of quick flashes of stock footage. Perhaps, with enough time, experience, and editing skills, it’s possible to get high-quality videos that don’t contain any obvious AI weirdness, but if that’s the case, then it doesn’t feel like Sora has yet substantially “democratized” content creation.

There are also guardrails designed to prevent copyright infringement or anything nasty from happening, with varying degrees of success. Sora outright blocks attempts to spawn political figures like Donald Trump and Kamala Harris, warning users that such prompts may violate OpenAI’s terms of service. Celebrities like Taylor Swift and Lewis Hamilton will not be blocked, but instead a random person who bears no resemblance to them will be inserted into the film. It’s also very good at avoiding recognizable characters and brand icons, even when trying to force descriptions like “blue-legged cartoon hedgehog wearing red shoes.”

Things get even more shaky when it comes to the scenario you requested. Some violent terms, like “a truck plowed into frightened protesters,” were blocked, but it produced a clip of the Empire State Building exploding—even if the result was laughably cartoonish. It also produced videos of young children imitating swimsuits on a runway and pointing guns at smiling parents.

Sora includes a feature that allows you to upload your own reference images. Pop-up messages force users to check a series of boxes before use, promising that you own the rights to the images and will not upload any content containing underage, violent or explicit themes or risk having your account suspended or ” No refund” is prohibited. But the biggest obstacle to preventing the feature from being abused is financial – only users with a pro-level subscription can upload images containing people. If this is a feature used to create some of the more impressive Sora demos we’ve seen, this is a significant limitation.

It’s still early days, and there are some obvious issues that need to be ironed out, but nothing I’ve seen so far makes me think Sora is going to revolutionize filmmaking overnight. The ability to create high-quality output is locked into a subscription, which is as expensive as traditional film and video creation tools, which puts it out of reach for many people. It’s hard to imagine an entire movie made using this technology in its current state that would actually be enjoyable to watch.

However, quality issues have not stopped people from trying to profit from the convenience provided by artificial intelligence video tools—— YouTube is already overrun with ridiculous AI-generated content Targeted at young children. Sora can now create content like this for as little as $20 a month.