Gen AI video trends: Where is all this going?

Current AI video examples which TV and movie producers should be aware of.

Hope you had a relaxing break over the festive holidays. As often is the case, a little period of downtime resulted in a new thought - and so here it is: a new semi-regular strand, to go alongside the creator case studies I’ve been sharing over the past year.

This is a little hard to quantify, but it is a broad catch-all to cover interesting and relevant trends, niches and audiences responses that are cropping up all over the place in relationship to gen AI as it becomes every more widespread in our lives.

It is less about the AI models and their advances - there already is a lot of noise about the latest version of Nano Banana or Kling (although if you want that too, let me know). Rather, each day there are instances where gen AI is used in ways that either directly or indirectly will have an impact on TV, film and online content markets. Say either as a new content trend with audiences, or instances which could be used within TV productions to bring costs down.

And perhaps unsurprisingly, a lot of it has much wider implications - thus explaining the title of this post (although more accurately perhaps it should be called ‘woah, where is all this going 🤯’). This reflects the multifaceted nature of what gen AI means to all of us. As a society, economies, individuals, producers, storytellers, creatives, and businesses.

Even if as a producer you make shows or movies far removed from social video/creator spaces, I still think this is important for all of us to be aware of these developments, not least as it might prompt a circling back to think about how this fits in with your development slate. After all, this content doesn’t exist for other people - it is appearing in all our feeds, and will increasingly appear in shows and professionally produced material.

In this post, I’ve covered:

• KlingAI’s new motion capture functionality that means anyone can become anyone based on a short bit of video and a still photograph as a reference

• Home renovation before and after timelapse and transition montages

• Never mind YouTube being TV, instead is it actually a replacement for daytime radio listening

• History and philosophy genres are being swamped by inaccurate and nihilistic content

• Natural disasters and tsunami content is now dominated by AI videos

• The volume of AI slop on YouTube, and how usage varies by country and demographic.

Before getting going, I thought I’d flag this sentiment that highlights how some people - often kids - like AI slop-like content:

my nephew is obsessed with ai shorts. cats playing instruments on people’s porches at night. people getting sucked up by tornados. talking animals. all kinds of zany stuff.

So while there are many deeply concerned about AI - around artistry, copyright, money, truth and so on - there are others that just think fantastical stunts and talking cats are hilarious.

Motion capture takes a leap forward

Kling AI has launched a new motion control version 2.6, which can take a short piece of video, and the using a still photograph recreates a video where the photo is matches to the actions of the video.

Here is an example where you can see the source material - a video of me moving my head and face around, and an old photo of Pamela Anderson grabbed from the internet.

It doesn’t quite look like how she looks in motion herself, however you can begin to imagine how this will develop over the coming months and years.

Here’s another one - another similar input video, and then a still of Tina Turner:

Here’s a short explainer demonstrating it in action:

Like everything with the internet, this has great potential but also profound risks (the copyright of the original images being just one). Already people are spotting the challenges around this type of technology - for example, why bother having real people when anyone could make themselves appear to be quite literally anyone (or anything - animal, vegetable and mineral)? And if you can fake this, then anyone who is accused of lying can say the footage was faked?

This video below demonstrates how a man could turn into a woman in real time - and in the original video which is here (X still doesn’t allow embedding of video content on Substack which is very annoying) his voice changes to a female’s voice too - with comments online saying that this type of functionality is ripe for use to catfish people online:

We are already seeing the application of this type of technology out in the wild, with an increasing number of AI generated hosted channels, for example:

If you look at some of the above videos, they are supposedly by John J. Mearsheimer, professor of political science at the University of Chicago. However the script appears to be AI generated as it deals in vague concepts, there is a lot of head and eye movements that don’t appear real.

Similarly the same channel has this video, supposedly featuring commentator George Will, however again it has tells it is AI generated - for example, mispronounces the phrase ‘live on air’, the background of books is very clearly fake, lots of eye blinking, and you can compare the real person here.

Watching these you can probably see the extraordinary potential for both good and harm this technology poses. These videos are risking the stoking of panic around geo-political hotspots, however right now, there are enough small tells for some people to spot that they aren’t real - soon, these signals will disappear, so how will we know if anything is true?

Considering the level of noise growing around this issue online, it does feel like something will be done about this - however what that is, how it happens, and who drives it forward is the huge question to watch over the next couple of years.

Home renovation shows: Makeover timelapse videos

AI is being used to generate versions of the classic timelapse transitions that has been a staple of home renovation, design and build TV shows for years.

Here is an example, where this person used Bytedance’s product Seedance Pro within RoboNeo AI. They explained their process, which in summary is (but the full process is here):

• generated a before and after image with the same perspective

• Use the AI tools first and last frame feature

• Use a detailed prompt outlining a timelapse of a home renovation.

And another user had a go using Kling, which is much more AI-like in feel and plausibility:

While all the material within this video is generated via AI, you could see how this type of tool could be used to make montages or transition timelapses within shows - so take a first shot, last shot, a few stills of your crew, and the time used to shoot and edit a timelapse sequence has been saved.

After the paywall there are more examples of gen AI trends - including a disturbing one for factual producers featuring a bunch of vikings and nuns…