In her second Ay? In the E-column, Julie Seal explains AI video through a bedside chat with her mother, carefully explaining the terminology around models, consistency, synthetic media, and tools that are reshaping ad production.
My sweet mom is in the hospital again and I’m holding a fort by her bedside, telling the nurses the names of complicated medications like I’m in the pit. It also means I don’t have as much time to immerse myself in work as much as I should (NHS Wi-Fi isn’t very useful for ‘generations’), but I also realized that I should probably start this AI video column with some Rory Stewart style commentary.
So for complex advertising concepts, or when I feel like I’m delving too deep into the flimsiness of the industry, I’m going to do it the way I’ve always done it. This is a “please describe me as if I were your Nan” test, and the “Nan” in question here is my mother Jean, whose actual first name is Barbara, confusing the aforementioned nurse.
Anyway, here we go.
myself: So Mom, did you know that I’m now using AI video for advertising?I thought I’d explain it a little bit on behalf of people in the industry who are too scared to listen and feel like they need to understand right away. Is that okay?
“Nan”: Who is Roxy?
myself: No one. Now, do you know how people can take photos and create videos for advertising? Now you can do it on your computer too. It’s almost always worse than the human method, but it’s also much faster and cheaper.
“Nan”: Okay…is that okay? Is it legal? I don’t think it’s legal. Don’t get into trouble.
myself: Yes, it’s legal, but you raised important points about ownership, copyright, ethics, and insurance. I have an upcoming interview with a lawyer who is an expert in the industry on this very topic.
“Nan”: Lawyer-san, Mr. Ju, what did you get yourself into this time?
myself: It’s okay, Mom. So there are a lot of different tools for generative AI, such as creating copy for ads, voiceovers, etc., and different “models” for creating images and videos. What is created is now sometimes known as synthetic media.
“Nan”: They look like an 80’s band with girly hair and baggy sleeves.
Want to know more? ask the drum
myself: There are some. So until now, the biggest problem when creating AI videos has been consistency. It’s about making the frames, scenes, characters, world, etc. the same throughout what you’re creating. I was making a video about pigs with lots of different shots from different angles and scenes. It was difficult to keep everything the same pig.
“Nan”: Oh, I love pigs.
myself: Well, these were “occupying” the center lane of the highway. I don’t even like puns. We were just trying to prove that something more complex and longer form could be done using a concept that was estimated to cost £200,000 to make in the pre-AI video world.
“Nan”: 200,000 pounds? You could even buy a house for it! That’s terrible, Julie. I hope you don’t spend £200,000 on a pig.
myself: No, it’s not, Mom. Anyway, there are things like “interpolation” where the AI creates “in-between” frames to make the transition between things look smoother, and “healing” which just changes small elements in the video, like hands or t-shirts. Then there’s “outpainting” which expands along the edges of the scene to create wide shots etc. There is also “compositing,” which overlays small parts of AI onto real-world footage. Or “temporal coherence.” The visuals and movements are stable over time and don’t shake or look strange. There’s a word called “hallucination,” which means that AI makes strange mistakes.
“Nan”: I’ve seen hallucinations!
myself: Yes, I know, Mom. That is fever. They’re giving you drugs for that. Anyway, what else can we talk about? Some of the current big names include Google Veo, Runway, Sora, Kling, and Pika.
“Nan”: Are they those? K-POP Demon Hunter Who is Betsy’s favorite character?
myself: Not completely. Google Veo is created by Google DeepMind. Runway is probably the most used advertisement currently. We also use a lot of models as kind of aggregator tools, and other platforms like Higgsfield use them as well. Sora comes from OpenAI. There’s a lot of competition now, but this model really made the industry take notice. Kling is made in China and is considered by many to be better than Western models. Pika tends to be popular with creators who create punchy social content and visual effects. All very cool. Oh, and there’s also Arcads for creating UGC-style “people chatting to camera on TikTok” stuff, but without actual humans. For audio, we have 11Labs, which has recently been improved with a variety of accents. They all sounded like Americans at one time. Oh, and I used Hedra for character stuff. It also became really popular.
“Nan”: That’s great, love. You might be able to take a nap now. You keep tweeting about tech stacks, consistency wars, and slop backlash, love. Don’t worry about it.
myself: mother? Already know about the hot topics in AI? What?
“Nan”: No, it’s probably just the painkillers affecting my brain. But I like to use the Chatty Bee Gees app I got on my phone, which tells me the names of plants. I can have him talk to me about anything. Judging by his voice, he’s a nice, very nice guy. And very smart.
myself: Oh, you’re definitely an AI guru. I love you, Mom.
“Nan”: I love you, Ju.
Read more opinions from julie. you can contact them here.
