OpenAI has published two new ‘Sora Showcase’ videos on YouTube made by professional creators. They show the impressive potential of the yet-to-be-released Sora AI video model.
Sora was first announced in February but due to concerns over time-to-generate, cost and misinformation potential, OpenAI has only made it available to a small group of creators.
Since its unveiling, two new Chinese AI companies have released ‘near-Sora-level’ models including MiniMax and Kling — at least Sora-level in terms of visual realism. More established startup Runway also released Gen-3 which comes very close to what we’ve seen of Sora.
Both of the new videos shared to YouTube demonstrate Sora's impressive character consistency and realistic motion. This is something we’ve seen from the start but is also something we’ve seen Runway, Luma Labs and others match with new models.
A focus on people and movement
One of the new Sora videos by Singaporean artist Niceaunties explores the concept of aging. The other, by British Korean artist David Sheldrick, explores fashion and modeling.
In both videos, there is a combination of complex motion, costume changes and a need for consistent characters. Also in both, we see that Sora can generate incredibly realistic-looking people, although there are some motion inconsistencies I haven’t seen from Sora before.
The problem with evaluating Sora’s capabilities in comparison to the readily available models is that we can’t see it at its worst. We don’t know how many terrible videos showing misformed limbs, merging bodies, and random hallucinations it makes for every good video.
So far Sora access has been restricted to filmmakers, artists and other creative professionals and all we’ve seen (for the most part) is the finished product after those failed attempts were discarded. In these new clips, we see hints of where Sora might struggle and it is in the same places as Runway Gen-3, Luma Labs Dream Machine or Kling — complex human movement.
Two new Sora videos
OpenAI has been sharing Sora showcase videos every few weeks showing off various aspects and capabilities of the AI video model as it starts to prepare for public launch later this year or early next year. I suspect the release will come after the U.S. Presidential Election.
The Niceaunties video is described as “an art project about aging, beauty, freedom & fun, and an attempt to understand ‘auntie culture’". The artist is an architectural designer by day and described Sora as "working alongside another human" producing incredible creations that "blur the line between reality and illusion" and make you question what is real.
The creator wrote: “What excites me the most is the ‘human-ness’ of the characters, particularly their facial expressions and the emotions they convey. There is immense potential for these visuals to forge strong emotional connections, making it an incredibly powerful communication tool. I also learn from the way Sora interprets prompts—the more I engage with it, the more I learn. I love that I am constantly learning while creating!”
For David Sheldrick, a graduate of the London College of Fashion and professional fashion photographer, it was about using Sora as a throwback to his early work as a backstage photographer at fashion shows, "a proving ground that set my trajectory for the rest of my life."
He wrote: “Sora is most powerful when creating new worlds & new physics. Just how real it can be, especially when creating humans, was really a ‘we have arrived’ moment. I have used a lot of different AI tools but nothing has yet come close to the realism of Sora.”
