OpenAI's highly anticipated Sora AI video generator has finally been released, but the results are far from satisfactory. Despite its impressive capabilities, Sora's generated videos are plagued by visual inconsistencies, distorted objects, and limitations that make it unsuitable for commercial use.
The AI video generator was released on Monday, after almost a year of teasers, and has already garnered significant attention. However, users who managed to sign up for the service were met with a slew of hurdles, including a subscription-based model that locks key features behind a $20 monthly "Plus" membership or a $200 monthly "Pro" subscription. The Plus tier allows users to generate videos at 480p or 720p, capped at either five or 10 seconds in length, while the Pro tier unlocks 1080p quality and 20-second-long videos.
Testing the Plus tier revealed underwhelming results, with simple prompts yielding the most realistic-looking videos. However, even these were not immune to visual issues, such as a cat being given a second tail for a few moments. More complex prompts resulted in glaring distortions, including hands flailing wildly when attempting to show someone applying makeup.
Sora's Storyboard feature, designed to help users lay out prompt instructions for longer videos, was also found to be lacking. The feature, which resembles a video editing timeline, allows users to explain what they want Sora to generate every two seconds. However, the results were even poorer, with added details resulting in more distortions and weirdness.
Despite its shortcomings, Sora did impress in certain areas. Video generation was faster than expected, taking under 30 seconds for even 10-second-long clips. Patterns on fur and textiles remained consistent, even during fast-paced movement, and the lighting, shadow, and mirror effects generated by Sora simulated real-world conditions well.
In comparison to other AI video generators, such as Runway AI, Sora's results looked more realistic and contained fewer visual distortions. However, the quality of Sora's outputs is on par with Adobe's Firefly Video Model, which has the added benefit of promising commercially safe generated outputs.
Despite its limitations, Sora is not without its potential uses. The AI video generator could be used to create quick flashes of stock footage, but it is far from being a viable replacement for traditional filming and video creation tools. The high cost of the Pro tier subscription also makes it inaccessible to many.
Furthermore, Sora's capabilities have already been exploited by individuals looking to profit from the convenience AI video tools provide. YouTube is already saturated with nonsensical AI-generated content targeted towards young children, and Sora is more than capable of churning out similar content.
In conclusion, while OpenAI's Sora AI video generator is an impressive first step, it falls short of expectations due to its visual inconsistencies, distorted objects, and limitations. Until these issues are ironed out, Sora is unlikely to revolutionize video production overnight.