OpenAI released Sora on Friday, 16th February. Watch the extraordinary generative AI content created by Sora here . There are nine impressive short sequences to watch covering a range of genre from complicated live action through animation; all generated by short text prompts. Has generative AI video content now reached a point where it is virtually indistinguishable from reality?  

David Shapton reports

Sora, OpenAI’s astonishing new text-to-video model for generating photorealistic videos is so good that most people don’t give it a second glance: they just assume that it is video. It really is that good.

It’s not incremental change; it’s a giant one. A year ago, we laughed at terrible pizza restaurant-type AI videos that were as terrifying as they were funny. Each frame looked like a different artist painted it, and any attempt to depict a human eating food was frankly grotesque.

But with Sora, if you’re not looking for AI artefacts, you won’t notice them. Crucially, there is no uncanny valley here. That’s a giant chasm to cross. The model also does physics. Light reflects, objects bounce off each other, fur moves in the wind, and waves lap like waves are supposed to lap. This looks like a decade’s improvement in less than a year – and that’s just on the surface. Does this mean it has an internal physics engine, just like Unreal or Unity? Nobody gave it one – it looks like it just emerged as a result of the complexity in its training data.

Under the hood, it’s clear that this software has a “world model”: it seems to “understand” the world. What does “understand” mean in this context? It’s hard to say, but it has passed a threshold where it can represent objects in the world and how they dynamically react to and with each other.

What does it all mean? Probably much more than any of us can know at this stage. How long before it affects us all? Film and video makers will be on the front line for disruption. But, beyond that – and I am far from the only one saying this – 2024 is a bumper election year. There are more elections in the next 300 days than in any year in history. There could not have been a more critical time to release software that can effortlessly deceive and do so plausibly.

I love technology. It’s thrilling to watch the extraordinary rate of progress. But we are close to losing the ability to predict what will happen next. The rate of change is so great that even experts in their own field are surprised (just google “emergent properties”). Sora may be the strongest evidence yet that we are in at least some kind of technological singularity.

Staff Reporter

Share this story

Share Televisual stories within your social media posts.
Be inclusive: Televisual.com is open access without the need to register.
Anyone and everyone can access this post with minimum fuss.