I’m usually supportive of new technology; however, I do have issues with some AI tools. While I’ve had some success with LLMs and image generation tools, video editing tools have been a nightmare. In this blog post, I reflect on my experience using AI avatars and outline some important lessons learnt.
When using AI tools, I’ve spent more time finding out they don’t do what I want, than if I had just done the work the old fashioned way. My newest gripe is with AI avatar tools. I’ve been asked on multiple occasions for my opinion on Synthesia, a tool that uses AI to generate videos with digital avatars. Basically, you write a script, and the avatar speaks it on screen. It’s often used for training, explainer, or presentation-style videos. While I love the concept, I think the execution is still a bit wooden. The faces still look a bit creepy, and the voices (especially British accents) are still robotic at times. However, I can see a place for this type of media when used in small doses.
To give a fair analysis, the Synthesia platform is very easy to use, and they have some really beautiful templates. I should also mention that I have also used HeyGen; the platform and usability are pretty much identical to Synthesia’s. What I should mention is that I didn’t finish creating a video in HeyGen, but the concept was pretty much the same. In addition, I want to be transparent about any bias and note that when I visited Synthesia at their stand at World of Learning earlier this year, they weren’t very welcoming and didn’t really want to engage with me as I didn’t represent a large company.
So what were the issues?
An issue that I didn’t expect to run into was the use of stock avatars. To provide context, I’m creating a series of videos which explore themes related to government policy and public finance. I’d spent quite a bit of time creating the first video using stock avatars only to be hit with an error when I went to generate the video.
Your video was moderated for violating our Deceptive Information & Other Restricted Content: Media Reporting (News) guidelines.
It turns out that there are quite a lot of restrictions on using stock avatars. This is because Synthesia wants to protect the individuals whose likenesses are used for stock avatars. They avoid associating those avatars with content that could be seen as political, controversial, or sensitive. In hindsight this makes sense, but I don’t think it’s very clear when you start using the software (or when you speak to a sales rep).
We are getting to the point where AI is talking to AI. An AI deemed my content to be related to government and a news story and, therefore, blocked it. A human then analysed this and judged it accordingly, but the AI and the human didn’t understand the context; they just identified keywords from my script. Upon reflection, I can totally see why they do this, but the process really lacks context.
At first, I couldn’t figure out why my content had been restricted and it was only when I purchased a paid for tier that I was able to access further information. I even used ChatGPT to try to revise my script to remove anything that could be a policy violation. After about an hour of speaking to an AI robot, I finally got through to a human who advised me that if I wanted to go any further with my particular content, I would need to use a personal avatar of my own likeness.
AI Birthing
I use the term AI Birthing to describe the creation of fictitious people using AI, like this lady on the left, who I created using ChatGPT. I had assumed that the models used in Synthesia were fake. I thought that AI had advanced enough to the point that the likeness of a person could be replicated. So if we want to protect the identity of models and actors, why are we still using real people for AI avatars? Surely this would circumvent a lot of the ethical politics around the use of someones image?
Lessons Learnt
I learnt a lot from this process. I should have conducted more investigation before simply signing up and using the product. I only found out that my content violated the policy once I had finished a video and went to generate it. If you’re on a tight budget and you’re looking for a tool where you want to use stock avatars to create video content (and not personal avatars), you must really think about the topic of your content and check the terms and conditions of the tool you’re looking to use.
Resources
If you want to check these tools out for yourself, check out the links below.