The landscape · Media and creative
Media and creative: what AI is really doing
Brief me in 2 minutes
AI in Media and creative.
AI is no longer a novelty bolted onto media, it is inside shipped, paid-for work: video shots in finished Netflix episodes, music trained directly on label catalogs under new contracts, and search answers that skip the article entirely. The fight has shifted from whether this is allowed to who gets paid for it and how much: labels and some publishers signed licensing deals, studios are suing over training data while writing AI pay rules into labor contracts, and the traffic AI search is taking from publishers is moving faster than the licensing money coming back.
The game: we show you 5 things happening. You call each one real now, early or mostly hype. Then the receipts.
Shift 1 of 5
AI video splits into model labs that train and aggregators that resell
Real now
Higgsfield says it is at $700 million in annualized revenue and Runway just closed a $315 million round at a $5.3 billion valuation. That is paying customers and real capital, not a pilot.
How it works. Runway and Moonvalley train their own video models and pitch them as future 'world models' for film and games. A second layer, led by Higgsfield, doesn't train a foundation model at all: it resells access to 15+ underlying models (Sora, Veo, Kling, Seedance) under one subscription and sells the workflow, templates and interface on top.
Overstated: "World models" is doing a lot of work in these pitches: none of these companies has shown a model that holds a character, set and story straight across a full scene, let alone a film. The revenue is real subscription money, the world-model claim is still a slide.
Shift 2 of 5
Consumer AI-video apps rose and died inside one product cycle
Mostly hype
The product that was supposed to prove video's mainstream AI moment is the one OpenAI itself killed within two quarters of launch.
How it works. OpenAI launched Sora as a standalone social app with a self-insertion 'cameo' feature, pitched as video's ChatGPT moment. Within seven months OpenAI shut the consumer app down, and the API follows on the same short runway, while rival labs (Google's Veo, Kuaishou's Kling, ByteDance's Seedance) kept shipping the same underlying capability inside other products.
Overstated: The 'ChatGPT moment for video' framing that greeted Sora's launch didn't survive two quarters. A capability going mainstream and a specific app winning are two different claims, and this is the case for treating them separately.
Shift 3 of 5
Music labels stopped suing generative AI and started licensing it
Real now
Two of three major labels have signed contracts, not press-release intentions, and Suno raised $400M on the strength of that legitimacy while still being sued by the third label.
How it works. Warner Music Group settled with Suno and signed a licensing deal instead of continuing to litigate; Universal Music Group did the same with Udio a month earlier and the two are building a jointly licensed AI music platform for 2026. Under the Warner deal, WMG artists can opt in to let Suno train on their voice, likeness and songs and get compensated for it.
Overstated: "Licensed AI music" branding papers over the fact that the unlicensed training already happened at scale before these settlements, and neither label has disclosed what individual artists actually get paid per opt-in.
Shift 4 of 5
Publishers are trading lawsuits for a licensing cut smaller than the traffic they're losing
Early
A handful of large publishers have a deal or a crawl fee; most of the web's publishers have neither one nor their old Google traffic back. This is a pilot response to an industry-wide loss.
How it works. AI search answers the query directly from crawled content instead of sending a click to the source. Ahrefs measured the effect getting worse, not better: a page ranking first now loses far more of its clicks when an AI Overview sits above it than it did a year earlier.
Overstated: Revenue-share programs like Comet Plus sound like the licensing problem is solved; they are opt-in, cover a short list of large publishers, and are paid out of a browser subscription most readers haven't bought.
Shift 5 of 5
Studios sue over AI training data while pricing AI performers into labor contracts instead of banning them
Real now
The SAG-AFTRA contract is ratified and in force, the Midjourney suit is in active discovery, and Netflix has already aired the AI-assisted shots in question. None of this is a proposal.
How it works. Disney, NBCUniversal and DreamWorks sued Midjourney in mid-2025 over generated images of Yoda, Darth Vader and other owned characters, and the case is now stuck in a discovery fight where Midjourney argues the studios' own internal AI use proves the industry treats this as fair use. In parallel, SAG-AFTRA's new contract doesn't ban synthetic performers, it prices them: any AI recreation of a performer must be paid the same as hiring the human, while Netflix has already used generative AI inside finished, aired episodes for shots like a crowd or a collapsing building.
Overstated: The SAG-AFTRA guardrail sounds protective, but the bar for using a synthetic performer over a human, 'significant additional value', is undefined and untested, and the union gave up its ability to strike over AI until the deal expires.
The scary part
What happens to jobs in Media and creative
What AI is already taking:
- generating VFX shots, like crowds or a building collapse, that a show would otherwise cut for budget TechCrunch ↗
- producing on-camera training and marketing videos without booking a presenter or a shoot HeyGen ↗
- being the answer, so the reader never clicks through to the article that actually reported it Ahrefs ↗
- training new commercial AI music models directly on a major label's catalog, with the label's sign-off Wikipedia ↗
What stays human:
Nobody has shipped an AI system that can direct a two-hour film, run a writers' room, or decide what a publication covers this week; every deal and lawsuit above is a fight over who owns and gets paid for the training data and the output, not over who makes the creative call. The legal liability for what ships, and the judgment about whether a shot or a track is good enough, still sits with a named human.
Get better · 15 minutes
Owning the parts of the process that carry legal and creative liability: rights clearance, continuity and quality control across an AI-assisted pipeline, and the judgment call on what a synthetic clip or performer is actually cleared to say, since that is exactly what SAG-AFTRA and the studios above are fighting to keep priced and permissioned rather than automated away.
Try this this week:
Rebuild one piece you already made, a training video, a b-roll shot, a jingle, using a named, available tool: HeyGen or Synthesia for a talking-head video, Moonvalley or Runway for a b-roll shot, Suno for a scratch track. Time the whole process including revisions, then compare that hour count and the tool's list price against what the same piece actually cost you last time.
Your move
The frontier model layer (Runway, Moonvalley, Google) is capital-intensive and already crowded. The open space is narrower: licensed-data models for buyers who need clean legal chain-of-title, and vertical tools that own one workflow end to end, the way Synthesia and HeyGen own corporate talking-head video, rather than another general text-to-video app competing with Higgsfield's aggregator.
Check whether the company trains its own model or resells someone else's under a subscription; that difference determines who has pricing power when the underlying model gets cheaper. Red flag: 'world model' language in a pitch with no long, controllable, continuous output shown, that is the same claim Runway is making at a $5.3 billion valuation without shipping one yet.
Put one piece of your content behind Cloudflare's pay-per-crawl or pay-per-use setting this quarter, then compare the AI-crawler revenue against the referral traffic Google's AI Overviews have already taken from that same page, using your own analytics rather than an industry average.
Prompt literacy is now table stakes and commoditized across a dozen models charging the same $10 to $30 a month. What isn't commoditized is directing an AI-assisted production: continuity control, rights clearance, and knowing when a synthetic shot or voice needs a real signed release. Start by producing one paid piece end to end with Moonvalley, HeyGen or ElevenLabs and tracking every hour it actually saved against your last traditional job.
What's actually changing
- 1
AI video splits into model labs that train and aggregators that resell
How it works. Runway and Moonvalley train their own video models and pitch them as future 'world models' for film and games. A second layer, led by Higgsfield, doesn't train a foundation model at all: it resells access to 15+ underlying models (Sora, Veo, Kling, Seedance) under one subscription and sells the workflow, templates and interface on top.
Why it matters. The aggregator owns the customer relationship and can switch suppliers overnight, while the model labs carry the compute bill. Pricing power in this market is landing with whoever owns the interface, not whoever trained the model.
What's overstated"World models" is doing a lot of work in these pitches: none of these companies has shown a model that holds a character, set and story straight across a full scene, let alone a film. The revenue is real subscription money, the world-model claim is still a slide.
RunwayHiggsfieldMoonvalleyLuma AI
3 receipts
- Runway raised $315M at a $5.3B valuation to keep pre-training larger models
AI video-generation startup Runway has raised a $315 million Series E round, nearly doubling its valuation to $5.3 billion
TechCrunch, 10 Feb 2026 ↗ - Higgsfield, which resells 15+ third-party video models under one subscription, hit $700M annualized revenue the same month it raised a $400M Series B
This month, the company also reached $700 million in annualized revenue.
PR Newswire, 17 Aug 2026 ↗ - Moonvalley raised $84M more (total $154M) with CAA money, built on video it says is entirely licensed rather than scraped
Moonvalley, the AI research company building foundational AI video models and tools trained exclusively on licensed content, announced it has raised $84 million in additional funding led by existing investor General Catalyst.
The AI Insider, 16 July 2025 ↗
- Runway raised $315M at a $5.3B valuation to keep pre-training larger models
- 2
Consumer AI-video apps rose and died inside one product cycle
How it works. OpenAI launched Sora as a standalone social app with a self-insertion 'cameo' feature, pitched as video's ChatGPT moment. Within seven months OpenAI shut the consumer app down, and the API follows on the same short runway, while rival labs (Google's Veo, Kuaishou's Kling, ByteDance's Seedance) kept shipping the same underlying capability inside other products.
Why it matters. The feed-and-cameo format didn't hold attention once every competitor could generate the same clip; the underlying capability survived, the standalone consumer app didn't.
What's overstatedThe 'ChatGPT moment for video' framing that greeted Sora's launch didn't survive two quarters. A capability going mainstream and a specific app winning are two different claims, and this is the case for treating them separately.
OpenAIGoogle DeepMindKuaishouByteDance
2 receipts
- OpenAI launched the Sora app with a feed and self-insertion cameos on September 30, 2025
Sora 2 was unveiled on September 30, 2025, with an iOS app at the same time
Wikipedia, 30 Sept 2025 ↗ - OpenAI shut the consumer app down seven months later, with the API scheduled to close right after this briefing's publish date
The Sora app was shut down on April 26, 2026, while the API is planned to be shut down on September 24, 2026.
Wikipedia, 26 Apr 2026 ↗
- OpenAI launched the Sora app with a feed and self-insertion cameos on September 30, 2025
- 3
Music labels stopped suing generative AI and started licensing it
How it works. Warner Music Group settled with Suno and signed a licensing deal instead of continuing to litigate; Universal Music Group did the same with Udio a month earlier and the two are building a jointly licensed AI music platform for 2026. Under the Warner deal, WMG artists can opt in to let Suno train on their voice, likeness and songs and get compensated for it. Sony is the holdout, still suing both companies.
Why it matters. The industry settled the fair-use question commercially rather than waiting for a court to rule: the price of forgiveness for training on a catalog without permission turned out to be a licensing deal plus artist opt-in, not a shutdown.
What's overstated"Licensed AI music" branding papers over the fact that the unlicensed training already happened at scale before these settlements, and neither label has disclosed what individual artists actually get paid per opt-in.
SunoUdioWarner Music GroupUniversal Music GroupSony Music
3 receipts
- Warner Music Group settled with Suno on November 26, 2025 and signed a licensing deal
Warner Music Group settled with Suno on November 26, 2025.
Dubspot, 26 Nov 2025 ↗ - Under that deal, Warner artists can opt in to have their voice, likeness and songs used to train Suno's models and get paid for it
Under the agreement, WMG artists and songwriters can opt in to have their names, images, likenesses, voices, and compositions used in AI-generated music on Suno, and are compensated for taking part.
Wikipedia, 26 Nov 2025 ↗ - Universal Music Group settled its Udio lawsuit on October 29, 2025 and the two are building a jointly licensed AI platform
In a landmark move late Wednesday (October 29) evening, Universal Music Group announced that it has settled its copyright infringement litigation against AI music platform Udio.
Music Business Worldwide, 29 Oct 2025 ↗
- Warner Music Group settled with Suno on November 26, 2025 and signed a licensing deal
- 4
Publishers are trading lawsuits for a licensing cut smaller than the traffic they're losing
How it works. AI search answers the query directly from crawled content instead of sending a click to the source. Ahrefs measured the effect getting worse, not better: a page ranking first now loses far more of its clicks when an AI Overview sits above it than it did a year earlier. In response, publishers are taking deals rather than only suing: Perplexity shares 80% of its Comet Plus subscription revenue with participating publishers, and Cloudflare now defaults to blocking AI crawlers on ad-supported pages unless the AI company pays.
Why it matters. The deals are real money for the publishers big enough to negotiate one, but they cover a slice of traffic while the click-through collapse is happening site-wide, including to publishers with no deal at all.
What's overstatedRevenue-share programs like Comet Plus sound like the licensing problem is solved; they are opt-in, cover a short list of large publishers, and are paid out of a browser subscription most readers haven't bought.
PerplexityGoogleCloudflareThe New York TimesNews Corp
4 receipts
- An AI Overview appearing above a search result now correlates with a 58% lower click-through rate to the top-ranking page, up from 34.5% a year earlier
the presence of an AI Overview now correlates with a 58% lower average clickthrough rate
Ahrefs, 4 Feb 2026 ↗ - Starting September 15, 2026, Cloudflare defaults to blocking AI crawlers on any page that carries ads unless the AI company pays
Starting on September 15, 2026, Cloudflare's default settings will block "mixed-use" crawlers from any pages that host ads
TechCrunch, 1 July 2026 ↗ - Perplexity keeps 20% of Comet Plus subscription revenue and pays the other 80% to publishers in its program
Perplexity keeps 20% of it, and the other 80% will go to participants in Perplexity's publisher program.
Digiday, 26 Aug 2025 ↗ - The year before, Ahrefs measured the drop at 34.5%.
Back in April 2025, we found that AI Overviews reduced clicks to top-ranking content by 34.5%
Ahrefs, 4 Feb 2026 ↗
- An AI Overview appearing above a search result now correlates with a 58% lower click-through rate to the top-ranking page, up from 34.5% a year earlier
- 5
Studios sue over AI training data while pricing AI performers into labor contracts instead of banning them
How it works. Disney, NBCUniversal and DreamWorks sued Midjourney in mid-2025 over generated images of Yoda, Darth Vader and other owned characters, and the case is now stuck in a discovery fight where Midjourney argues the studios' own internal AI use proves the industry treats this as fair use. In parallel, SAG-AFTRA's new contract doesn't ban synthetic performers, it prices them: any AI recreation of a performer must be paid the same as hiring the human, while Netflix has already used generative AI inside finished, aired episodes for shots like a crowd or a collapsing building.
Why it matters. Studios are running two tracks at once: litigate the input (whether training on their IP was theft) while normalizing the output (AI-generated shots and performances shipping in real productions under negotiated pay rules). Either way, AI-made content is already on screen.
What's overstatedThe SAG-AFTRA guardrail sounds protective, but the bar for using a synthetic performer over a human, 'significant additional value', is undefined and untested, and the union gave up its ability to strike over AI until the deal expires.
DisneyNBCUniversalDreamWorksMidjourneySAG-AFTRANetflix
3 receipts
- Disney, NBCUniversal and DreamWorks sued Midjourney in mid-2025, showing the AI reproducing copyrighted characters like Yoda from simple prompts
the AI is willing to produce fairly realistic depictions of Star Wars character Yoda in response to the prompt
Georgetown Law Technology Institute, 24 June 2025 ↗ - Midjourney is arguing the studios' own AI use is evidence the industry itself treats this training as fair use
if the studios "are themselves training generative-AI models on third-party copyrighted works, scraping the internet and investing in tools that do the same", this would serve as "powerful evidence that the relevant industry considers such conduct to be fair use"
The Art Newspaper, 9 July 2026 ↗ - SAG-AFTRA's new studio deal lets AI 'synthetic performers' replace an actor only when they add significant additional value to the production
Producers committed to “a principle strongly favoring human performances” and agreed not to deploy AI-generated “synthetic performers” — digital characters with no real-world counterpart — unless doing so offers “significant additional value” to the production.
Gazettely, 12 May 2026 ↗
- Disney, NBCUniversal and DreamWorks sued Midjourney in mid-2025, showing the AI reproducing copyrighted characters like Yoda from simple prompts
Who's doing it
11 companies to know, 7 with a round we can source. Newest first where dated.
Search these companies in the startup database ↗ · Explore all research tools
Higgsfieldvideo and image generation, model aggregator
Subscription platform that bundles 15+ third-party video and image models into one product for creators, brands and agencies instead of training its own frontier model.
Latest · 17 Aug 2026Reached $700M in annualized revenue the same month it closed a $400M Series B at a $5.4B valuation PR Newswire ↗
$400M Series B, 17 Aug 2026 source ↗
HeyGenAI avatar video
Identity-first AI video platform that generates talking-head presenter video for marketing and training without booking a person or a shoot.
Latest · 25 June 2026Crossed $200M in annual recurring revenue, doubling in eight months HeyGen ↗
SunoAI music generation
Generates full songs, including vocals, from text prompts; settled with one major label while still fighting two others in court.
Latest · 3 June 2026Raised a $400M Series D at a $5.4B valuation while still facing active copyright lawsuits TechCrunch ↗
$400M Series D, 3 June 2026 source ↗
Runwayvideo generation, world models
Trains its own text and image-to-video models and is pushing toward general-purpose 'world models' for film, games and simulation.
Latest · 10 Feb 2026Raised a $315M Series E at a $5.3B valuation to pre-train larger world models TechCrunch ↗
$315M Series E, 10 Feb 2026 source ↗
ElevenLabsvoice generation and cloning
Text-to-speech and voice-cloning platform used for dubbing, narration and voice agents, now a top-tier funded AI voice company.
Latest · 4 Feb 2026Raised $500M in a round led by Sequoia at an $11B valuation, more than tripling its prior valuation TechCrunch ↗
$500M Series D, 4 Feb 2026 source ↗
SynthesiaAI avatar video, corporate training
AI avatar video platform enterprises use to make training and product videos without a camera crew or on-camera talent.
Latest · 26 Jan 2026Raised a $200M Series E at a $4B valuation, nearly doubling its valuation from a year earlier TechCrunch ↗
$200M Series E, 26 Jan 2026 source ↗
Black Forest LabsAI image and video models
Makes the Flux family of image and video generation models, licensed to and embedded in tools like Krea, Canva and Freepik/Magnific.
Latest · 1 Dec 2025Raised a $300M Series B at a $3.25B valuation, bringing total funding past $450M TechFundingNews ↗
$300M Series B, 1 Dec 2025 source ↗
UdioAI music generation
AI song-generation platform that settled its copyright suit with Universal Music Group and is building a jointly licensed AI music platform with the label.
Latest · 29 Oct 2025Settled its lawsuit with Universal Music Group on October 29, 2025 and agreed to a new licensed platform launching in 2026 Music Business Worldwide ↗
PerplexityAI search, publisher licensing
AI answer engine and Comet browser that answers queries directly from indexed content and shares subscription revenue with participating publishers.
Latest · 26 Aug 2025Comet Plus splits subscription revenue 80/20 in favor of the publishers whose content is served in answers Digiday ↗
Moonvalleyvideo generation, licensed training data
Builds its own video model, Marey, trained only on licensed footage rather than scraped web video, targeting studios that need clean legal chain-of-title.
Latest · 16 July 2025Raised $84M more (total $154M) from CAA, CoreWeave and Comcast Ventures to expand its licensed-footage library VP Land ↗
$84M Strategic extension, 16 July 2025 source ↗
MidjourneyAI image generation
Text-to-image generator that built a large subscription business with no outside venture funding, now being sued by three major studios over training data.
Latest · 24 June 2025Being sued by Disney, NBCUniversal and DreamWorks since mid-2025 over Midjourney reproducing their copyrighted characters from simple prompts Georgetown Law Technology Institute ↗
Your move
Our read, built on the receipts above.
The frontier model layer (Runway, Moonvalley, Google) is capital-intensive and already crowded. The open space is narrower: licensed-data models for buyers who need clean legal chain-of-title, and vertical tools that own one workflow end to end, the way Synthesia and HeyGen own corporate talking-head video, rather than another general text-to-video app competing with Higgsfield's aggregator.
Check whether the company trains its own model or resells someone else's under a subscription; that difference determines who has pricing power when the underlying model gets cheaper. Red flag: 'world model' language in a pitch with no long, controllable, continuous output shown, that is the same claim Runway is making at a $5.3 billion valuation without shipping one yet.
Put one piece of your content behind Cloudflare's pay-per-crawl or pay-per-use setting this quarter, then compare the AI-crawler revenue against the referral traffic Google's AI Overviews have already taken from that same page, using your own analytics rather than an industry average.
Prompt literacy is now table stakes and commoditized across a dozen models charging the same $10 to $30 a month. What isn't commoditized is directing an AI-assisted production: continuity control, rights clearance, and knowing when a synthetic shot or voice needs a real signed release. Start by producing one paid piece end to end with Moonvalley, HeyGen or ElevenLabs and tracking every hour it actually saved against your last traditional job.
What's next
- 24 Sept 2026 OpenAI's Sora API is scheduled to fully shut down, following the consumer app it already closed in April en.wikipedia.org ↗
Other industries
- FinanceBanks move agentic AI from chatbots to running back-office and dealmaking work
- Software engineeringSpaceX buys the leading AI coding startup outright, for $60 billion
- HealthcareAmbient scribes go from pilot project to clinical infrastructure
- LegalLegal AI vendors go from research assistant to agent that produces the deliverable
- Real estateLeasing and resident calls move to voice AI agents that never sleep
- Sales and marketingCustomer support agents now get paid per resolved ticket, not per seat
- Jobs, HR and recruitingAI runs the first-round interview, not just the resume filter
- InsuranceClaims move from adjuster-first to AI-first, humans supervise the exceptions
- CybersecurityAutonomous AI agents take over penetration testing and vulnerability hunting
- Consulting and accountingAI-written client reports become a recurring, public Big Four embarrassment
- Retail and e-commerceShopping agents get a checkout button, and Amazon spends 2026 deciding who's allowed to press it
- EducationNYC and LA ban AI for students, not for teachers
- Manufacturing and roboticsHumanoid robots move from demo livestreams to paid warehouse shifts
- Transport and robotaxisRobotaxis go from demo to metered fare across a dozen-plus US cities
- Energy and data-centre powerGrid rules get rewritten mid-buildout, state by state
- GovernmentGSA turns AI adoption into a government-wide subscription, then flips 'free' to metered
- Agriculture and foodAutonomous tractor fleets learn to run the tools behind them, not just the wheel