Steerability, so fine-grained control of video, will get better. For image generation, AI is already able to review and reflect on what it sees, then adjust only where needed or according to instructions. Sometimes it takes several tries, but it works. Video is harder, of course, but likely doable.
I find this plausible. A lot of people use nano banana to help maintain consistent characters in the resulting videos (this is part of what I was talking about with storyboarding), and I would not be surprised if a lot of those tools end up being made in video as well.
Though I will note that image generation is still not perfectly steerable! This is an awesome breakdown: https://genai-showdown.specr.net/
Sora 2 already does this, it just doesn't do it well. Occasionally, you'll see some element that doesn't look like typical AI video generation at all, and more like a bad attempt at tracking in a VFX element. This usually happens when the thing that's being generated doesn't have any real world references, and the steering AI notices the poor result and "manually" fixes it.
The tech is undoubtedly cool... the effects of this tech on society much less so. Advertising is already 90% slop and with the cost barrier to entry being effectively taken down, are we just going to get more slop? At least in the old slop world, actors and videographers and editors and etc were still being paid... Not trying to rain on this parade, the tech is mind blowing. I just don't know if we're ready for this as a society. Interesting read, thank you!
FWIW, I tried hard to not editorialize about the technology when writing this. I think it's important and interesting that these major brands are moving toward AI advertising, and I didn't want to cloud the piece with my own opinions.
Certainly I think that there needs to be more discussion on whether this is a good development and how to steer towards better futures (e.g., I think I'd support labeling standards for AI generated media, though ofc a lot of details to work out there).
I suspect a bigger transition in AI advertising will come at the point where it’s possible to generate unique ads for particular viewers. Somewhere I saw someone mention the possibility of clothing ads featuring someone who looks like the viewer (using their Sora2 profile info) wearing the clothes, but you could also imagine having ads feature characters strolling around streets or parks where the viewer spends a lot of time (as indicated by their phone location history) or other subtle transformations of a moderately standardized ad. And if you’re in a context where prompt-and-go ads are good enough, the prompt could be customized based on the viewer’s targeting data.
Couldn’t you stage, frame and animate in a tool like Blender using simple consistent models/animatics then export for refined final output to AI? Instead of a gamble each time you get enough controlnet rigour to force the result you want?
I think, realistically, we could be three to four years away to where every ad you see on television is created by AI,” said Kavan Cardoza, a filmmaker who co-founded an AI-based studio, PhantomX.
That really puts it into perspective. New worlds soon.
Great callouts. Is there a list of vendor watermarks...SynthID - what else?
I wonder if there's a (small, micro-niche) market for a web browser extension that detects and flags these on the fly. Why, yes...yes, I believe there is.
Steerability, so fine-grained control of video, will get better. For image generation, AI is already able to review and reflect on what it sees, then adjust only where needed or according to instructions. Sometimes it takes several tries, but it works. Video is harder, of course, but likely doable.
I find this plausible. A lot of people use nano banana to help maintain consistent characters in the resulting videos (this is part of what I was talking about with storyboarding), and I would not be surprised if a lot of those tools end up being made in video as well.
Though I will note that image generation is still not perfectly steerable! This is an awesome breakdown: https://genai-showdown.specr.net/
Sora 2 already does this, it just doesn't do it well. Occasionally, you'll see some element that doesn't look like typical AI video generation at all, and more like a bad attempt at tracking in a VFX element. This usually happens when the thing that's being generated doesn't have any real world references, and the steering AI notices the poor result and "manually" fixes it.
The tech is undoubtedly cool... the effects of this tech on society much less so. Advertising is already 90% slop and with the cost barrier to entry being effectively taken down, are we just going to get more slop? At least in the old slop world, actors and videographers and editors and etc were still being paid... Not trying to rain on this parade, the tech is mind blowing. I just don't know if we're ready for this as a society. Interesting read, thank you!
FWIW, I tried hard to not editorialize about the technology when writing this. I think it's important and interesting that these major brands are moving toward AI advertising, and I didn't want to cloud the piece with my own opinions.
Certainly I think that there needs to be more discussion on whether this is a good development and how to steer towards better futures (e.g., I think I'd support labeling standards for AI generated media, though ofc a lot of details to work out there).
Excellent piece, after which I feel I learned something beyond what someone thinks I should be annoyed about.
I suspect a bigger transition in AI advertising will come at the point where it’s possible to generate unique ads for particular viewers. Somewhere I saw someone mention the possibility of clothing ads featuring someone who looks like the viewer (using their Sora2 profile info) wearing the clothes, but you could also imagine having ads feature characters strolling around streets or parks where the viewer spends a lot of time (as indicated by their phone location history) or other subtle transformations of a moderately standardized ad. And if you’re in a context where prompt-and-go ads are good enough, the prompt could be customized based on the viewer’s targeting data.
Amusing idea, you should raise $50 million and do it.
Sounds alarmingly plausible.
Awesome read! highly informative
Couldn’t you stage, frame and animate in a tool like Blender using simple consistent models/animatics then export for refined final output to AI? Instead of a gamble each time you get enough controlnet rigour to force the result you want?
this shows the new phase of AI adoption — when even global brands treat AI like a standard creative tool, not a gimmick.
now it’s down to how fast teams can adapt their workflow.
the same shift hit writing, design, and consulting before: first denial, then scale.
what matters now is who learns to brief AI well enough to keep the human taste in the loop.
I think, realistically, we could be three to four years away to where every ad you see on television is created by AI,” said Kavan Cardoza, a filmmaker who co-founded an AI-based studio, PhantomX.
That really puts it into perspective. New worlds soon.
Great callouts. Is there a list of vendor watermarks...SynthID - what else?
I wonder if there's a (small, micro-niche) market for a web browser extension that detects and flags these on the fly. Why, yes...yes, I believe there is.
The pace of change is wild. We’ve gone from “AI can write copy” to “AI can make full ad films” in what feels like months.
But it’s interesting to see how teams keep the soul of storytelling intact when creation is this instant.
I couldn't stop reading, so beautifully written, great examples and research! ✨