Demand for Sora catalytic computing power continues to soar, further benefiting infrastructure … [+]
The world is getting a sneak peek at a major transformative technology of our time – one that will likely displace more resources than some of the other media models we’ve seen in the past.
With the ability to create entire videos from prompts, Sora is set to make all kinds of casting, directing and expensive film work obsolete. And it’s supposed to be available right now! At least to ChatGPT Plus and Pro users.
A big change
Think of all the work involved in making, using and developing physical film for pictures. Consider what the digital camera has done to this industry.
Sora is going to be so much more transformative, at such a scale, that it’s really worth paying close attention to as it gets into the hands of a wider group of users.
Marques Brownlee is a fairly influential YouTuber who got a front row seat to early Sora exploration.
In a YouTube review, Brownlee points out how capable and tailored the program is at creating realistic footage, while admitting it still has a ways to go.
Was it Gandhi or Stalin?
For those who have been playing the game where you guess which person said what, Brownlee shows us a series of short videos and asks us to figure out if they are real or created by AI.
What you see is that it can be very difficult to tell.
There isn’t a lot of glaring evidence to tell us which videos are AI-generated and which aren’t
Not many clues
There is some obvious information about AI video, but it is mostly based on our knowledge of the world around us.
Here are three that popped up for me:
It’s not real – one way you can tell a video is created by AI is if it actually has the wrong elements in an image, such as a landscape you’ve seen before.
If you know, for example, that there isn’t a dilapidated shack on top of some hill, you’ll notice the difference when the AI shows you an aerial view.
Not quite “ugly” – it seems that Sora products are still following the same trend as still images and general AI technologies in general – they tend to produce shiny, beautiful results. In other words, one of the few things it tells you is that the program doesn’t tend to create average-looking film with real composition or lighting problems, or subjects that aren’t telegenic and film-ready.
However, you could probably create this sort of thing with more prompts.
Not reliable – another way to tell AI videos is if you see tentacle monsters or other unreal things appear on the sidelines. But again, this has more to do with our knowledge of the world than anything we can assess visually. If you see a tentacle god emerging from a lake in a Sora video, it will look real.
Now, after writing these, I watched the rest of Brownlee’s video and he mentions some additional mistakes in Sora hyperrealism, so check them out:
First, there is the lack of item permanence, where items or characters can appear and disappear randomly. There is also a kind of “ghost image” effect where an object will lack substance – for example, in Brownlee’s video, we see cars driving through another car in what is supposed to be a real street video.
In general, Brownlee points out, Sora struggles with physics. It doesn’t always know how moving objects behave or what direction they were going if it’s working on a still image.
Then there are some speed issues where a video can be slowed down and sped up for no reason.
On top of all that, some of the videos will be so realistic that we have a hard time distinguishing them from live footage.
Access Sora Right now
On OpenAI’s website, it appears that access to Sora is temporarily down due to high traffic volume.
Brownlee talks about this possibility in his video:
“I wonder how long it will take when it’s open for everyone to use,” he says, noting that a roughly 10-second 1080P movie takes a few minutes to create.
Use cases for Sora
As Brownlee points out, Sora may be more useful to many people who want to create animation or slow motion features.
This is because physics are really hard to get right in real life, but animation and stop action footage are more forgiving. They’re more abstract, and this will be one of the first areas where Sora becomes more useful, although Brownlee also mentions fake CCTV camera footage as a desirable way to use the platform.
It shows a fully AI-made “Santa versus Frosty” Mortal Kombat gameplay video…and a job interview scene in which Sora shows off some of the details the model can provide without any extra prompting.
However, he suggests there are some big unknowns with this technology and we’re moving into uncharted waters.
When we all start using Sora, what will we use it for? How will it affect our lives?
People are wondering what is going to happen to these big industries that are often concentrated in places in Southern California like Burbank and Hollywood.
But the effect of technology will likely go far beyond that. Stay tuned as I continue to document what is now coming to the end of what has been a banner year for major language models.
