There’s no doubt that AI will have a big role to play with video production but the quality of tool varies.
I’m a big fan of the CapCut editing app and news of a ChatGPT plug-in got me excited.
Here’s what I did with using the tool.
I’m from Staffordshire. So, I decided to make a video for Staffordshire Day which actually happened a couple of weeks back.
The prompt I gave it was to include the landmark of Stafford Castle as well as the greatest British footballer of all time Sir Stanlety Matthews. I wanted a Stoke City reference and the Staffordshire knot. Drive into the county and you’ll see the roadsign welcoming you to Staffordshire the Creative County so I wanted a reference to that.
It produced a ascript for the film.
It also asked me what aspect ratio I’d like the film to be. 16:9 is landscape and 9:16 is portrait. I wanted portrait. I also decided to keep the script as it was.
What the video looked like
It makes sense for ChatGPT to synch with a video creation tool like CapCut. It means that you are able to use the tool to edit and replace elements you are not happy with. However, CapCut does need a minimal level of professional expertise. You can change the images that has been created. You can also do this editing on a laptop or PC.
How good was AI shot selection? Ropey, frankly
Okay, now to the business end. I have to tell you that the choice of images was not great. AI chose around the images I asked for. If you squint you’ll say the images corresponded with what it was asked for by and large. However, knowing Stafford only two of the images were accurate and just one – the shot of Stafford Castle – would pass muster.
The first shot was a shot of a Herefordshire sign (wrong), an image of a horse statue (wrong) a statue that isn’t Stanley Matthews and one that is. Two Stoke fans looking dejected (hmmm) and two landmarks (wrong). The last shot was a red and while image with the word osteoperosis (spectacularly wrong).
Here’s the video
Despite being told the video was 9:16 the images were more landscape 16:9…
As you’ll see if you played the video, the voiceover was in a pretty horrible American accent and the music selection feels pretty lift music.
What worked and what didn’t?
What did work was the process. I asked ChatGPT to produce something and it did.
But the images chosen were poor.
What’s worrying is that the images were copyrighted.
Using tineye I could see the Stafford Castle image was listed on Shutterstock, Alamy and Adobe.
The Stoke City clown was also a Shutterstock stock image.
The statue wasn’t Sir Stan at all but is from Amsterdam.
So, in a crude way, the process works but the images selected here are a hospital pass. Rights holders like Getty are going after those who use images without permission. As ever, if rights holders think you’ve got budget they’ll come after you.
You can export the video and you have some control over the frame rate it can export at. That’s a plus.
What did potentially work for the process is the script. It did produce a workable script after a prompt listing the features. I wouldn’t be using the provided audio commentary. I’d be using someone with a Staffordshire accent if I was making this in earnest.
I’m sure that the results will improve as time ticks on but the ChatGPT CapCut extension isn’t worth using just yet.