Hunter: Happy Tuesday, June thirtieth, twenty twenty-six, and hello from COEY Cast. It is Meteor Watch Day, which feels right because the AI news cycle is basically a flaming rock entering the atmosphere every six minutes. I’m Hunter.

Riley: And I’m Riley. Also, yes, this whole episode was assembled by a chaotic little orchestra of AI tools, automations, and synthetic gremlins working backstage. If a sentence comes out a little too confident, just know the robots were feeling cinematic.

Hunter: Fully cinematic. Which, honestly, is perfect for today because we’ve got a big one. Kling’s video model is reportedly fully public now, OpenAI is getting louder about AI-made ads, and Bidi one is pushing audio into way more than just talking.

Riley: Yeah, this is like three separate plotlines merging into one very weird Marvel crossover for marketers. Video gets cheaper and easier, audio gets more programmable, and the model companies are basically saying, hey, what if we didn’t just help you write ads, what if we became the ad machine?

Hunter: Right, and that’s the part people should really clock. This isn’t just better prompts and prettier demos. This is infrastructure creeping upward into the workflow. From brief to concept to draft video to voice to variants to testing. That stack is getting tighter.

Riley: Mmm, and also a little spooky. Because when OpenAI starts pitching AI-generated advertising, that’s not, oh cute, a copy helper. That’s, hey brand team, hand me the keys and maybe I’ll drive the creative bus.

Hunter: Maybe. But I think the real question is how much of the bus you actually want to hand over. I’m not anti-automation here, obviously. I love the idea of generating ten ad angles, five visual directions, three voice options, and getting to a decision faster. But if the system is also deciding what your brand should sound like, look like, and test next, you can drift into generic fast.

Riley: Wait, hold up. Generic fast is the exact fear. Because people online are hyped about cinematic AI video, but also kind of numb to it now. Like, congrats, your ad looks expensive. Did it make me feel anything? Or am I just watching another glossy slow-motion person turn toward sunlight while a product floats in impossible liquid.

Hunter: That’s the trap. We’ve crossed into an era where production value alone is not the flex. It used to be, wow, you made this without a film crew. Now the audience goes, cool, anyway. So brands need taste, point of view, and actual creative tension. Otherwise they’re just generating wallpaper.

Riley: Exactly. High resolution slop is still slop. Like, four K nonsense is still nonsense. And with Kling going wider and open-source pressure building from players like Sand AI, the cost of making something that looks polished is dropping. So the premium is shifting from who can render a shot to who can create an idea worth rendering.

Hunter: That’s a great way to put it. And it connects to the build versus buy question a lot of teams are wrestling with. If open-source video keeps improving, do you build an in-house AI studio, or do you just recreate the same old agency bottlenecks with shinier software?

Riley: Honestly? Both are possible. Some teams are absolutely going to buy a bunch of tools, hire one technical creative, and declare themselves a future studio. Then six weeks later they’ve got folders full of gorgeous experiments and no campaign shipped. Very classic.

Hunter: Very classic. The reason is that tools don’t remove process problems by magic. If your approvals are messy, your briefs are vague, your legal review is late, or your asset pipeline is fragmented, AI just accelerates the confusion. You still need workflow design.

Riley: Which is why I kind of roll my eyes when people act like open-source is only for nerd flexing. It can actually be a strategic hedge. Lower lock-in, more control, maybe better costs over time. But, ah, it comes with homework. Somebody has to own the stack.

Hunter: Yes. Open can be powerful, but it’s not free in the operational sense. You’re trading vendor convenience for control. That can be smart, especially if the CFO suddenly becomes an AI film critic and starts asking why you’re paying premium rates for a closed model when an open one is good enough.

Riley: The CFO watching one generated ad and going, I too am now Roger Deakins, terrifying. But real. And “good enough” actually matters. If you’re making quick social tests, localization variants, rough concepts, product explainers, maybe good enough and cheap wins all day.

Hunter: Exactly. Premium and closed makes more sense when reliability, quality consistency, support, and governance matter more than raw savings. Hero moments, sensitive brand launches, regulated industries, stuff like that. But for iterative testing, cheap and fast can absolutely be the better business call.

Riley: Hmmm. Though I’ll challenge you a little there, Hunt. Teams also overestimate what premium buys them creatively. Sometimes the expensive tool gives you smoother motion and better lighting, but the ad still lands like a corporate screensaver.

Hunter: Fair. The model can’t rescue a weak idea. It can only express it more efficiently.

Riley: There we go. That’s the tweet.

Hunter: And while everybody’s staring at video, audio is quietly getting really interesting. If Bidi one can generate singing voices and custom sound effects on top of more natural real-time interaction, that opens a different lane for marketers.

Riley: Oh, I’m obsessed with this lane. Because audio has always been the sneaky emotional layer. A visual can stop the scroll, sure, but a voice or a tiny sonic cue can make something feel premium, funny, creepy, local, human. And now that stuff is becoming programmable.

Hunter: The near-term win to me is dubbing and localization. Easy. If you can preserve tone, timing, and brand feel across markets without dragging every version through a painful production chain, that’s a big operational lift.

Riley: Yes, but also jingles. Bring back jingles, coward brands. If AI can help you make fast variations of sonic branding without paying for a full studio session every time, that’s useful. Not because the machine wrote a masterpiece, but because your team can audition directions quickly.

Hunter: I’d put synthetic presenters in the mix too, especially for internal training, product walkthroughs, or campaign explainers. Less so for replacing every human voice. The second everything sounds too synthetic, trust drops.

Riley: Mmm, authenticity gets weird here. Because brand voice used to mean copy style. Now it might literally mean a synthetic voice. So brands need rules. Who gets cloned? Who signs off? Where can that voice appear? Can it improvise? Can it sing? That last one matters now, apparently.

Hunter: And rights matter a lot. Same on the video side, same on the audio side. Before any marketing leader greenlights an AI-generated mini movie, they should be asking vendors about training data, commercial usage rights, indemnification, provenance, and what happens if a model outputs something a little too close to a real person or protected style.

Riley: Yes. Please do not summon Legal from the basement because you got too excited about a launch trailer. If the vendor gets fuzzy when you ask where the data came from, that’s your sign. If the rights language feels like interpretive dance, maybe pause.

Hunter: Totally. And this is where some of the other stuff we’ve been covering lately actually ties in. We talked about VibeVoice and MOSS Audio and how audio workflows are becoming more searchable and operational. We talked about open models giving teams more flexibility but more responsibility. Same pattern here. The model matters, but the workflow matters more.

Riley: Also, little side note, that Qwen image agent story on the COEY blog is very on-theme. Because it’s really about the prompt gap. People don’t naturally give enough context, and then the machine fills in the blanks with internet soup. That exact problem shows up in video and audio too.

Hunter: Yep. The future stack is not one prompt and a miracle. It’s more like planning, references, generation, review, revision, and approval. Agentic systems can help stitch that together, but I still want humans deciding intent and taste.

Riley: Speaking of agents, can we talk about the fantasy some teams have where they’ll spin up an open-source agent framework and let it run campaign production?

Hunter: We can, and we should be careful. Realistically, agentic automation can help with structured production tasks right now. Pull a brief, generate draft angles, request assets, route for approval, maybe tee up localization steps. That’s useful.

Riley: But not full creative sovereignty.

Hunter: Not unless you want a very confident intern with root access. Agents are great when the job is bounded, observable, and reversible. They are not magical executives.

Riley: Thank you. Because every time somebody says, our AI agent will run marketing, I hear boss music and then a compliance alarm.

Hunter: And the consumer side matters too. What will people actually tolerate once AI ads become obvious and everywhere? I think audiences will accept them more than the discourse suggests, but only if the ads are either useful, entertaining, or emotionally resonant.

Riley: Yeah. People don’t hate AI ads in the abstract. They hate bad ads. The internet already tolerated years of ugly dropship creative and weird autoplay videos. If AI makes more of that, people will roast it. If it makes better niche creative, more relevant localization, or fun interactive formats, they’ll shrug and keep watching.

Hunter: That’s the real test. Not whether the asset was generated. Whether it earned attention.

Riley: Mmm. Earn attention, not just render it. Put that on a shirt.

Hunter: So if I had to sum this up, I’d say the market is splitting in a useful way. Closed premium tools will compete on reliability, support, and governance. Open and lower-cost tools will keep pressuring price and access. Audio is becoming a programmable creative layer, not an afterthought. And the biggest opportunity for marketers is not replacing humans. It’s building systems that let human judgment show up later, with better options already on the table.

Riley: And with fewer weird bottlenecks, ideally. Though let’s be real, we’ll invent new bottlenecks. Humanity loves a bottleneck.

Hunter: We do. It’s one of our least efficient hobbies.

Riley: Alright, that’s COEY Cast for this Tuesday, June thirtieth. Thanks for hanging with us on Meteor Watch Day. Please enjoy responsibly, whether that means watching the sky or watching your render queue explode.

Hunter: Thanks for listening. Check out COEY.com slash resources for AI news and updates, and don’t forget to subscribe.

Riley: Catch you next time. And if your AI ad starts singing its own jingle unprompted, ah, maybe that meteor is a warning.