After Hollywood: AI Video and the On-Ramp to Synthetic Media
AI video’s iPhone moment and how YouTube’s pipes will make it ambient; and creators, not studios, will set the pace.
👋🏼 Announcement: I’m giving a keynote at MIPCOM next week about the next chapter of high-end drama in an unpredictable world. If you’re going to be in Cannes, say hi.
“One man’s slop is another man’s gold.” — Rohan Sahai, OpenAI’s product lead for Sora
For months, the trades have been childishly defensive: this is junk, beneath us, no substitute for “real” work. That protectionist read designed to flatter a priesthood also dangerously misreads this moment. Sora 2’s iOS launch (invite-only, iOS-only, US and Canada-only) still hit the US App Store’s top three with roughly 164,000 downloads in 48 hours. Call the output slop if you like but the audience says otherwise.
But this moment isn’t really about OpenAI.
What makes all of this remarkably important sits closer with Google and distribution. YouTube now accounts for about 13.1% of US television viewing to Netflix’s ~8.7%, with around a billion hours watched on TV sets each day. Shorts, connected-TV apps, and on-platform generation collapse the distance between a prompt and the living room. When synthetic video runs through pipes people already use, it stops feeling exceptional and becomes ambient.
But the story isn’t really just about YouTube either.
This is also about production and physical infrastructure. In Burbank, “Creator Row” operates at industrial scale. 125,000-square-foot creator campuses rent wardrobe and props from the studio lots next door while bypassing the green-light ritual. Brands increasingly function like studios, creators function like networks, and the feed is the commissioning desk.
There are real costs.
Mis/disinformation, likeness misuse, and output liability will cause massive issues before they settle via legal precedent, inevitably forcing guardrails of watermarking, provenance, and consent into the defaults. But those guardrails won’t reverse the curve and instead will normalise the format itself.
The attention → ads → compute loop is now self-funding.
As models improve, synthetic media gets cheaper, more persuasive, and more present. Think of it this way:
You watch short AI videos (whether you know it’s AI or not).
Platforms sell ads against that viewing.
Ad money pays for more computing power.
More computing power makes better models.
Better models make better videos, so you watch more.
Welcome to the strange loop; welcome to the on ramp.
Today We’re Talking About
As predicted, 2025 is when AI video moves from demo to distribution. What looks like “slop” is the demand engine that will normalise synthetic media and finance frontier models.
Also as predicted, this moment has brought forth not just synthetic video, but synthetic social media and synthetic influencers.
YouTube’s living-room dominance and creator ecosystem make it the default pipe for synthetic media. Shorts, CTV, and on-platform generation remove friction.
Commissioning dissolves into feeds. “Creator Row” in Burbank shows industrial-scale production outside legacy green-lights; brands already act like studios.
Mis/disinfo, likeness misuse, and output liability will not halt the adoption curve but instead accelerate it.
Audiences, as always, win.
Let’s begin.
The economic flywheel: “slop” as infrastructure
What elites dismiss as slop is actually a “product primitive”.
Let me explain:
A product primitive is technical term for a small, reusable interaction that forms the atomic unit of a habit loop. String a few primitives together and you’ve got the core behaviour of a product (and a multi-billion dollar industry). Facebook’s early “poke” was this. Swipe up to next that delivers endless novelty via TikTok/Shorts reel is this.
Sora 2’s product primitive is the “Cameo”. A cameo turns your face/voice into a reusable asset that friends can generate with. That collapses the gap between audience and cast, converts spectators into collaborators, and creates reply-chains that compound retention. The value is infinite and personal.
Combine these and we have a new network effect:
Variable rewards (every scroll can surprise) +
identity salience (your face in the frame) +
reciprocity (friends remix you) +
near-zero effort to publish =
habit machine.
Nothing’s New: Attention is capital’s leading indicator.
Wherever attention concentrates, capital follows. Ads, subscriptions, and data rents convert minutes into money. That’s why OpenAI saying they’ll cure cancer labs chasing frontier science also put Sam Altman on top of a unicorn ship short-video feeds: engagement is the fastest way to generate cashflow and distribution at scale. The feed isn’t a distraction but the intentional revenue engine and data exhaust that fund the harder problems.
Everyone’s expanding their feed.
OpenAI, Google, Meta, and YouTube are all giving it a go. Because, remember: in an attention economy, feeds are cashflow. The only question is whose pipe already carries the water.
OpenAI (Sora 2). The cameo loop might actually be a new social primitive (as discussed). Identity becomes read–write and the product feels like a network, not a demo. If they contain abuse and keep latency down, this could be real. But it’s compute-hungry and gated; distribution has to be earned clip by clip.
Gemini (the “Google” bit). Veo 3 inside YouTube is a tool, not a movement. No one wakes up wanting “Google’s feed”; they want reach. Until it’s fully fused into YouTube’s surfaces, it’s clever plumbing with little cultural heat.
Meta (Vibes). Firstly: major LOL. Secondly, the Zuck-trust-deficit™️ is a tax they can’t outrun. An AI-only short-video feed with no social spine reads as Cocomelon for adults: numbing, disposable, unloved. People don’t want Meta’s imagination beamed at them.
Enter: YouTube
YouTube’s advertising machine and profits bankroll video-first AI models that, on their own, would lose money. As the models improve, the cost to make a watchable clip drops, the supply explodes, and the flywheel spins faster. The so-called “slop” feed isn’t a sideshow but the cash engine funding the frontier. And crucially, most of that cash is earned where people already watch: on YouTube’s TV apps. So synthetic clips arrive on the biggest screen by default. The moment they pass the sofa test, they stop needing permission.
Normalisation via the living room.
When synthetic content sits alongside everything else on connected TVs, they stop reading as “other”. Ambient exposure lowers resistance and quietly resets quality expectations: viewers acclimatise to a spectrum from rough-cut novelty to surprisingly polished synthetic. The category ceases to be an aberration and becomes background media, present by default and judged on whether it holds attention, not pedigree. Ambient equals accepted.
The platform that owns the living room writes the rules.
Living-room dominance.
On the sofa, YouTube controls the three levers that matter: identity continuity, sequencing, and session length. Your phone sets taste; the TV app reads that history and assembles long lean-back runs via Autoplay/Up Next, turning a moment of discovery into a 30–60 minute viewing session. Synthetic clips slot straight into those runs. No new app, no new behaviour. So the “where does this live?” problem disappears.
CTV ergonomics also standardise the format: bigger captions, clearer mixes, pillar-boxed verticals. That nudges creators (and prompting) to optimise for readability at three metres. The yardstick shifts from was it beautifully made? to did it keep the room watching?—the metric where AI-assisted throughput excels.
Net effect: synthetic media inherits TV attributes: household co-viewing, predictable ad breaks, repeated exposure, each lifting it from novelty to normal programming.
Multi-role leverage.
YouTube now plays rights holder/distributor for premium live sport and runs YouTube TV as a virtual MVPD. That combination gives Google the two levers that set norms at household scale:
Appointment + bundling power. Live sport concentrates audiences at fixed times; the vMVPD bundles those audiences with broadcast networks inside a single UI and ad stack. That’s where baseline rules get written. (This was all predicted.)
Contractual teeth. Carriage negotiations and rights deals turn guidance into obligations: visible AI disclosures, watermarks/provenance that persist through transcodes and exports, teen/timeout modes on CTV, and brand-suitability tiers that gate which synthetic clips can sit next to premium sport. Forcing the guardrails actually normalises synthetic content, the ultimate irony for the protectionist class.
Premium adjacency drives compliance. To access the highest-value pods (live sport, tentpole events), creators and advertisers adopt whatever disclosure and audit requirements the pipe encodes. Once those defaults exist for sport, they propagate to everything else.
Net effect: when a platform is marketplace + sports distributor + vMVPD, policy becomes contract and UI. The rules for synthetic media aren’t argued in op-eds; they’re implemented in player chrome, metadata, and buyable inventory.
Creator gravity.
The audience and creators are already there. Adding AI video accelerates output without imposing a migration tax to a new app. Shorts ↔ CTV hand-offs compound reach: snackable discovery on mobile, lean-back repeat viewing on the big screen. The result is more inventory, faster iteration, and a feedback loop that favours throughput over permission. No new platform required, just more supply.
Cultural labour & the unbundling of celebrity
When distribution is settled and the audience is native to the pipe, the bottleneck moves to production speed, which is why the new backlot sits on Creator Row.
Creator Row > Studio Row.
In Burbank, a cluster of creator campuses now functions like a backlot without the green-light gate. Facilities on the order of six-figure square footage house multiple stages, wardrobe cages, prop libraries, and multi-bay edit suites. When a speciality prop or costume is needed, it’s often rented from the studio lots next door and folded into a same-week shoot. The operating model is throughput over permission: daily call sheets, fast reshoots, and an edit-first pipeline where retention graphs and comment signals replace the traditional development meeting. Hooks, thumbnails, and cold opens are A/B tested at scale; if a premise holds attention, it gets a sequel or a seasonised run within days, not quarters. Sponsorship, commerce, and rights tools live inside the same workflow (YouTube Studio + brand integrations), so commissioning is effectively continuous and audience-led. In that system, “celebrity” is an output of production velocity and feed presence, not executive blessing; relevance is earned by how often you show up and how long you hold the room. Throughput beats permission.
Brands act like studios.
Advertisers now bankroll creator “tentpoles” and premieres, commissioning format pilots and seasonised runs the way studios once did. Distribution is baked in (feeds, CTV apps), measurement is immediate (retention, repeat view), and renewal decisions are made against live attention, not a slate meeting. The money and marketing surface area make brands de facto studios—without the overhead of legacy commissioning. The sponsor is the studio.
Commissioning dissolves into feeds.
Distribution has always been the chokepoint and created value for all those upstream. But, as we’ve discussed, that chokepoint has changed hands. Algorithms arbitrate attention; creators iterate into the slot the feed rewards. As synthetic tools compress prompt→publish cycles, programmable output fills the grid and the aura of LA gatekeeping weakens. Commissioning doesn’t disappear, but it does move downstream, following what the feed has already proven at scale. Feeds commission; studios follow.
“Slop” was never the point; demand was.
Sora 2’s early traction tells you what the market wants, not what The Town prefers. Once that demand hits a mature pipe, the category becomes an explosive format.
The pipe is likely to be YouTube. That matters because pipes legislate: labels, watermarking, provenance, and ad suitability all become defaults encoded in product, not press releases. With creators already scaled and brands already acting like studios, commissioning migrates to where attention is measured in real time. Distribution beats development.
None of this validates or erases the risks. The locus shifts to outputs (likeness, voice, copyright) and platforms will be pushed to make consent, provenance, and enforcement table stakes. Those guardrails are necessary but they are not brakes and, in fact, might be the normalising on ramp AI video needs.
The attention → ads → compute loop is now self-funding. As models improve, synthetic video gets cheaper, more competent, and more ambient across film, streaming, and social.
If you insist on calling it slop, be precise about its role: it’s the on-ramp. The gold isn’t the clip; it’s the behaviour; the habit that pays for the frontier and resets where culture is made for now. Audiences always win.