Creative Intelligence Feed
Tracking the model releases, production shifts, legal issues, and research breakthroughs reshaping filmmaking, branded content, and visual storytelling.
We watch the AI production landscape in real time — generative video and image models, Hollywood adoption, legal and provenance shifts, and research breakthroughs — and translate it into practical signal. The news is everywhere; the TTI Takeaway is why it matters for your work.
An R-rated AI feature, ‘Gods Don't Give Gifts,’ heads to theaters
YouTuber Zack London (Gossip Goblin) released the trailer for Gods Don’t Give Gifts, a feature made entirely with AI, opening in limited release in the US and UK on December 4. The filmmakers say it is the first AI film to receive an MPA rating — an R — and to be campaigned for Best Animated Feature.
The rating and the Oscar run are the story, not the technology. An MPA rating and a qualifying theatrical window mean the existing system is processing an AI film as a film. The audience came from a YouTube channel, not a studio — which is a distribution lesson as much as a production one.
Google's Genie 4 world model adds audio, longer worlds, and real-time steering
Google DeepMind released Genie 4 as a research preview, generating interactive worlds with real-time music, ambience, and dialogue, and holding a scene consistent for up to two minutes — double Genie 3. Users can change weather, camera angle, or add objects mid-simulation, with Gemini planning and checking the steps.
Two minutes of persistence is the number that matters, not the audio. That is roughly the length of a scene, and it is where world models stop being demos and start being usable as a virtual set you can return to between takes. Google lists drift over long simulations as a known gap — which is the exact thing a director would test first.
Anthropic sued over Elvis Presley recordings used to train Claude
Authentic Brands Group and other Presley rights holders sued Anthropic in federal court in San Francisco, alleging 274 songs were used without permission to train Claude and that the model reproduced them on request. They seek statutory damages of up to $150,000 per song and an injunction; Anthropic said it disagrees and will defend the case.
The training-data fight is no longer confined to image and video models. Estates with a single, famous catalogue are a different kind of plaintiff from a label group — narrow claims, easy to explain to a jury. For anyone producing music-led work, it is another reason to ask what every model in the chain was trained on, including the ones that only write.
Advertisers pull Meta ads after unwanted AI “glow-ups”
Meta's updated commercial terms make advertisers solely responsible for anything its AI features create or change in their ads — including features switched on by default. Bookshop.org found an AI illustration added to a human-made ad, and agency Exverus pulled live ads after unwanted AI edits.
The finished master is no longer the last version of an ad. If the platform can regenerate a frame after delivery, approval and clearance end at the upload, not at the screen. Brands should check which AI features are on by default in every ad account — and agencies should write that check into delivery.
Canva releases Craft, its first image generation model
Canva Craft plans a layout before rendering it, outputs up to 2K in 30 aspect ratios, and was trained on human-designed Canva templates rather than photographs. Results can be split into layered, editable designs, and the model is available in Canva's Dream Lab and API.
Editable layers are the real feature. A flat generated image is a dead end the moment a client wants the headline moved; a layered one can enter a normal design revision cycle. Expect video tools to be judged the same way — not on how good the output looks, but on how much of it you can still change.
Mise-en-Video: staging multi-shot video generation in 3D
A multi-shot text-to-video framework in which each shot is staged first with oriented 3D boxes for subjects, then generated with attention that keeps the story consistent across shots. The authors built a 160,000-video dataset of multi-shot scripts with 3D annotations and report gains over baselines in consistency and subject control.
The name says it: blocking before generating. Research is converging on what previs has always done — decide where things are in space, then make the image. The tools that win in production will be the ones that accept a staged scene as input, not just a paragraph.
BBC chief says he has watched an AI-generated ‘Doctor Who’ episode
BBC director-general Matt Brittin, who previously ran Google EMEA, said someone had sent him a 50-minute, entirely AI-generated Doctor Who episode, called it “pretty good,” and declined to criticise it.
A fan-made 50-minute episode of a broadcaster's biggest brand is now a thing that exists, and the head of that broadcaster has seen it. The real friction is not whether it can be made, but who is allowed to say it is good — and a broadcaster’s chief praising an AI version of its biggest brand puts that question in front of everyone who works on the real one.
AMD to acquire Fei-Fei Li's World Labs for $8.2 billion
AMD agreed to buy World Labs, maker of the 3D world model Marble, in an all-stock deal worth $8.2 billion. The deal is expected to close by year-end, subject to regulatory approval.
A chipmaker paying this much for a world-model lab says where the compute demand is expected to come from next: persistent, explorable 3D, not just clips. For productions, the near-term effect is consolidation — fewer independent tools, more of them tied to a hardware platform.
COVER: video watermarks that survive compression
An invisible video watermark that hides its payload in the latent space of a frozen generative video autoencoder and is trained against simulated codec damage. The authors report 93.72% average bit accuracy across four codecs at twelve settings.
Provenance that does not survive a re-encode is not provenance — and every delivery, upload, and social cutdown is a re-encode. With the EU labelling rules now in force, marking methods that hold up through real codecs move from research curiosity to compliance requirement.
Danny Boyle confirms generative AI in his new film ‘Ink’
Danny Boyle told Vanity Fair he used around 30 seconds of generative AI in Ink, his film about the launch of The Sun newspaper, which opened the Venice Film Festival before playing Telluride.
Compare this with Eli Roth a month earlier: same scale of use, opposite reception. Boyle said it himself, in an interview, before anyone had to find it. Disclosure is becoming the difference between a craft decision and a scandal — and it costs nothing to get right.
Runway introduces GWM Worlds 2, a real-time interactive world model
Runway's research preview streams continuous 720p video at 24fps with 48kHz audio, with no preset session length, steered by text actions and continuous camera movement. Runway lists its own limits: detail degrades on fast camera turns and long-term memory is imperfect.
24fps and 48kHz are film and post numbers, not gaming ones — Runway is pitching this at people who deliver to picture. The stated weak spot, fast camera moves, is the one that matters most for action, so treat it as a tool for locations and mood today, not for coverage.
LightBridge: relighting a Gaussian splat in a single pass
A feed-forward method that relights an entire 3D Gaussian Splatting scene in one pass, without per-scene optimisation, trained on a large multi-illumination dataset. It targets the baked-in lighting that limits how splat captures can be reused.
Splats are the fastest way to capture a real location, and baked lighting is what keeps them from matching anything shot on another day. Relighting without per-scene training is the step that turns a scan into a usable set extension.
Universal and Sony amend their Suno lawsuit to add stream-ripping claims
A US judge allowed the labels to add claims that Suno ripped recordings from YouTube in breach of anti-circumvention law, but refused to expand the case from 560 named tracks to more than 61,000. The labels filed the amended complaint on August 25.
The argument is shifting from whether training is fair use to how the data was obtained. That is a much easier case to make, and it applies to video models trained on scraped platforms just as much as to music.
MSEditor: one edit, applied consistently across multiple shots
A framework for applying the same edit across several shots that differ in angle, scale, and subject pose, trained on repurposed multi-view video. The authors report better identity preservation and temporal stability than existing editing methods on their own benchmark.
Editing one shot is a demo; editing a scene is a job. Change the costume, the prop, or the time of day, and it has to change in every angle of the coverage. Cross-shot consistency is the gap between generative editing and a real notes round.
The MPA and ByteDance agree IP protections for Seedance and Seedream
The Motion Picture Association and ByteDance signed a memorandum of understanding on IP guardrails for the Seedance video and Seedream image models across TikTok, CapCut, and Dreamina. It follows the MPA's February cease-and-desist over Seedance 2.0; the specific safeguards and enforcement terms were not disclosed.
The studios are moving from suing model makers to negotiating with them — at least the ones with consumer platforms to protect. Undisclosed terms mean nobody outside can check them, so for client work the practical rule does not change: guardrails in the model are not clearance for the output.
Higgsfield raises $400M, quadrupling its valuation to $5.4B in eight months
The AI video app maker raised a $400 million Series B led by DST Global at a $5.4 billion valuation, up from $1.3 billion eight months earlier. Higgsfield reports $700 million in annualised revenue and 30 million users.
Higgsfield sells finished workflows — cinema and marketing studios — rather than raw model access, and it is reporting $700 million a year doing it. The money in AI video is moving to the interface layer, which is also where creative control is won or lost.
Addressable memory lets video world models return to earlier scenes
NVIDIA and Princeton researchers show that interactive world models lose track of earlier scenes once a rollout runs longer than their training length, and propose a fix that needs no retraining. They report 15.5% better temporal consistency and 19.5% better recall on a new benchmark that tests returning to a place after a long detour.
Walking away and coming back is what a camera does all day. A world model that forgets the room when you turn around cannot hold continuity, so this is the specific problem to watch — and it is good to see it measured directly.
Alibaba unveils Wan3.0 with 30-second clips and mixed inputs
Wan3.0, released in beta, doubles maximum clip length to 30 seconds and accepts text, image, video, and audio inputs together, alongside documents such as PDFs and slides. It can also suggest a clip length from the prompt and extend existing videos.
Thirty seconds is a full spot, not a shot. Accepting a deck or a PDF as input is aimed squarely at marketing teams generating from the brief itself — which makes the brief, not the prompt, the thing that needs to be well written.
Eli Roth admits ‘Ice Cream Man’ features AI-assisted shots
After suggesting he had drawn an animated sequence in Ice Cream Man himself, Eli Roth said days before release that AI was used in a few scenes, adding that he had “misspoke.” Social media posts had pointed to the involvement of an AI VFX studio.
The use was small; the denial was the problem. Audiences now go looking for exactly this, and they found it within days. Any production using generative work should decide its disclosure line before the press tour, not during it.
Black Forest Labs launches FLUX 3 Video
The Freiburg lab's first video model is generally available through its API, producing clips of up to 20 seconds with native audio and lip-synced dialogue in more than 14 languages. It supports keyframes, continuation from existing footage, multi-shot clips, and a cheaper draft mode.
A draft mode priced well below final is the feature that maps to how production actually spends: explore cheap, commit expensive. It is also the most serious European video model yet, which matters to clients who care where their tools and data sit.
EU AI Act labelling rules for AI-generated content take effect
From August 2, the AI Act's transparency rules require AI providers to mark generated output with watermarks or machine-readable markers, and deployers to disclose deepfakes. Clearly artistic, creative, satirical, or fictional work has lighter disclosure duties, and systems already on the market have until December 2 to comply.
The creative exemption is narrower than it sounds: it covers how a work is disclosed, not whether the underlying files are marked. Advertising is not fiction. Any campaign running in Europe should now treat provenance metadata as part of the deliverable, and agencies should know which of their tools write it.
German court rules against Suno in landmark GEMA training-data case
The Munich Regional Court found that Suno breached German and US copyright law by training on GEMA's repertoire, and held that AI companies must license material for systematic use and commercial exploitation.
For anyone producing for the European market this is the most consequential ruling of the year. It moves training-data provenance from a reputational question to a licensing obligation, and German and EU clients will start asking which models were used and on what basis. Answering that requires records kept during production, not reconstructed afterwards.
TARS: re-shooting existing video from a new camera angle
A method that regenerates an existing video from a new viewpoint chosen by prompt — a different shot size, angle, or first- versus third-person — without 3D reconstruction or paired multi-camera footage. The authors report more accurate camera control than prior methods, including reverse-angle shots.
A reverse angle you never shot is the most requested thing in an edit and the hardest thing to fake. Research that generates it from a single take points at coverage becoming something post can add, not just something the shoot has to deliver.
Runway launches a model router as generative media crowds
Runway launched routing that sends a request to whichever model suits it, rather than committing every task to one in-house model — a response to a field where no single model leads at everything.
This is tiered model strategy becoming a product feature. We have argued for a while that committing everything to one tool costs more and produces less predictable results; a major vendor now building routing into the platform is that argument conceded from the other side.
Netflix used generative AI on roughly 300 titles this year
Netflix disclosed in its Q2 earnings that around 300 titles across its library used generative AI somewhere in production this year — spanning concept and previsualization through post-production and release, rather than concentrated in any single stage.
The number that matters is not 300, it is the spread across the pipeline. Adoption is not arriving as AI-generated shows; it is arriving as AI inside conventional production at every stage. That is exactly the shape of demand a hybrid pipeline serves, and it is why buyers now ask vendors how AI fits their existing process rather than whether it does.
Meta launches Muse Image, and users push back over use of their photos
Meta released Muse Image, a free AI image generator in the Meta AI app, Instagram Stories, and WhatsApp. One feature lets users tag anyone with a public Instagram profile and turn that person’s photos into new AI images; the owner is not notified and has to switch it off in settings.
The consent argument has left the courts and reached the feed. Until now the consent fight was fought by rights societies and publishers — GEMA, the studios, the news outlets. Here it is about ordinary people’s faces, used by strangers, with opt-out as the default, and users objected on their own behalf. For anyone commissioning work, it is a reminder to ask early not just what a tool can produce, but whose likeness it can pull in without asking.
A definition and roadmap for world models
An attempt to pin down what a world model actually is and what would count as progress, at a point where the term is being applied to very different systems by different labs.
Useful precisely because "world model" has become a marketing term. If you are being sold one, this is the vocabulary to ask questions with — persistence, controllability, and whether the thing holds together when you move the camera.
Midjourney moves to expose studios' own AI use in copyright defence
Defending against an infringement suit from Disney, Universal, and Warner Bros., Midjourney is seeking discovery into how those studios themselves use AI, arguing fair use and pointing to the plaintiffs' own practices.
Whatever the outcome, discovery is the story. Studios are being asked to produce records of their own AI usage, which raises the internal cost of undocumented generative work across the industry. Expect documentation requirements to tighten in vendor contracts well before any judgment lands.
Google invests $75M in A24 to build AI filmmaking tools
Google put $75 million into A24 to co-develop AI-powered filmmaking tools — a model provider taking an equity position in a studio known for director-led work rather than licensing to it at arm's length.
The interesting part is the partner. A24 is the studio whose brand rests on taste and authorship, which is precisely the objection most often raised against generative tools. If the tooling that comes out of this is built to serve a director's intent rather than to replace it, it sets the standard everyone else gets measured against.
DGA deal lands health and residual gains alongside AI-era terms
The Directors Guild settled a contract covering health plan gains and increases in residuals, and responded to the television slump by limiting actors and others moving into directing jobs.
The guilds are negotiating AI terms against a backdrop of contracting television work, which sharpens every position. Read guild outcomes as a signal about volume as much as about technology — the two pressures are being felt at the same table.
SAG-AFTRA ratifies a four-year deal with synthetic-performer terms
Members ratified a four-year contract with the major studios covering synthetic performers. Producers may use an AI performer only where it brings significant additional value over a live actor or that actor's digital avatar.
"Significant additional value" is the phrase to internalise. It makes a synthetic performer a creative justification you have to argue, not a budget line you can quietly choose. Anyone scoping union work now needs that argument written down before the shoot, not assembled afterwards.
Scorsese adopts AI for storyboarding, and what it signals
Variety weighed what Martin Scorsese’s adoption of AI — limited, for now, to storyboarding — means for filmmaking, coming from one of the medium’s most prominent traditionalist voices.
Sentiment among established filmmakers is the gate on high-end adoption, and it is moving. This matters commercially: when directors of that standing engage rather than dismiss, the conversation with talent shifts from whether to use these tools to who can be trusted to use them well.
Real-time AI video startup Reactor raises $59M
Reactor, founded by former Apple engineers, raised $59 million from Jeffrey Katzenberg's WndrCo and others to build real-time AI video generation.
Real-time is the threshold that changes the creative process rather than just the cost line. Generation fast enough to sit inside a review session turns iteration into direction — you stop submitting prompts and start giving notes. That is a different craft, and it favours people who can actually give notes.
Amazon MGM greenlights three AI series and launches a purpose-built production platform
Amazon MGM Studios greenlit three animated series for Prime Video under a new GenAI Creators' Fund, alongside Project Nara — a production platform built with AWS that integrates AI agents with Maya, Blender, Nuke, Unreal Engine, and Adobe tools.
Note what it integrates with. A studio building an AI platform around Maya, Nuke, and Unreal is conceding that generative work has to enter the existing DCC and review pipeline, not replace it. Vendors who can deliver into those formats and that review cadence stay relevant; vendors who only export finished MP4s do not.
Runway's studio deals put generative video inside real pipelines
A profile of Runway's move from filmmaker tool to infrastructure, including signed deals with Lionsgate and AMC Networks and its position against Google's Veo and Genie.
Named studio deals are the adoption signal worth tracking, because they survive procurement, legal, and delivery specs in a way that demos never do. When Lionsgate signs, the question inside every other studio changes from whether to who.
Inside the AI protections in SAG-AFTRA's studio deal
A clause-level read of the new agreement's AI provisions, alongside the annual minimum increases and the merged pension plan.
Worth reading properly rather than in summary if you produce anything union-covered. The obligations that bite in practice are the consent and disclosure mechanics, and those are the parts that never survive a headline.
Studios and SAG-AFTRA confirm the new agreement
The AMPTP and SAG-AFTRA confirmed terms on the new contract, closing a negotiation in which AI provisions were the most contested element.
The deal date matters for scheduling: productions crossing it are subject to the new AI terms. If you are quoting work that straddles the boundary, confirm which contract governs before you price the synthetic elements.
Runway's CEO argues AI video was the prequel to world models
Cristóbal Valenzuela makes the case that text-to-video is a transitional capability, and that simulating a persistent world with consistent physics is the actual target.
If this is right, the craft shifts from prompting a shot to directing inside a scene — blocking, camera, and continuity become available again rather than being things you fight the model for. That is a return to something much closer to conventional filmmaking, which is good news for people who already know how to do it.
Obsidian Studio details its AI filmmaking approach
The Hollywood Reporter profiled Obsidian Studio, the AI production company partnered with Ron Howard and Brian Grazer's Imagine Entertainment, on how it is applying generative tools to film development and production.
The competitive pattern is now clear: AI production companies are winning legitimacy through partnership with established production names rather than through technology claims. Provenance of people, not just of pixels, is what unlocks studio-level work.
Phantom: physics-infused video generation
A method that jointly models visual appearance and latent physical dynamics during generation, rather than treating physical plausibility as something that emerges by accident from enough training data.
Physics is where generated footage still gives itself away — cloth, water, hair, and anything with weight. Approaches that model dynamics explicitly rather than hoping for them are the ones worth watching, because that failure mode is exactly what forces a shot back into comp.
India's Abundantia and InVideo launch an AI film studio
Abundantia Entertainment and InVideo announced a slate of five AI-driven films over three years, operating through Abundantia's AI division aiON using InVideo's generative video technology.
AI-native production companies are forming fastest in markets where the gap between ambition and traditional budget is widest. Expect the strongest early proof of what these pipelines can carry to come from outside the US studio system, not inside it.
Runway raises $315M at a $5.3B valuation, aiming at world models
Runway raised $315 million at a $5.3 billion valuation, explicitly to push beyond video generation into more capable world models.
Read the stated goal, not the number. The company closest to filmmakers is betting that clip generation is a waypoint and spatial, persistent scene understanding is the destination. That is the difference between generating a shot and being able to move the camera in it.
SAG-AFTRA weighs a studio tax on digital performers
Unable to block synthetic performers outright, the union is considering a levy on studios that use them — shifting the argument from prohibition to pricing.
This is the shape most of these fights will settle into: not whether synthetic performers are allowed, but what they cost and who is compensated. Budget for that outcome rather than for a ban.
Runway ships a world model and adds native audio to Gen-4.5
Runway released GWM-1, a world model working by frame-by-frame prediction with an internal sense of physics, alongside a Gen-4.5 update bringing native audio and one-minute multi-shot generation with character consistency.
Multi-shot generation with character consistency is the capability that turns a demo into a scene. Consistency across cuts has been the single biggest reason generative work stalls before the edit — the part our pipeline has had to solve with references, comp, and cleanup. Every increment the models make here is work that moves upstream.
Tencent forecasts AI in up to 30% of long-form film and animation
Tencent projected that 10 to 30 percent of long-form film and animation could be dominated by or deeply involve AI within two years, framing generative tools as a supply-side shift in premium screen content.
Treat the number as directional rather than precise. The useful signal is that a major platform is planning capacity around it, and animation is named first — the discipline where generative work has the shortest path to finished frames and the fewest likeness and clearance problems.
PhyWorldBench: benchmarking physical realism in text-to-video
A benchmark evaluating how well text-to-video models obey physical laws, covering a broad range of scenarios rather than scoring on general visual quality.
Benchmarks decide what gets optimised. As long as models are ranked on how good a still frame looks, that is what improves. Measuring physical realism directly is how the field starts fixing the thing that actually costs production time.
Europe's dubbing industry leads the fight over AI voice
European dubbing actors have secured AI clauses in national agreements — Italy's ANAD contract prevents cloning or re-use of a voice performance without explicit consent and compensation, and is being used as a template in Spain.
Localisation is where AI hits European production first, and the terms are being set territory by territory rather than centrally. If you deliver into multiple European markets, voice rights are now a per-country question — and the answers are diverging.
Turn AI signals into production systems.
Trained to Imagine helps studios, brands, and creative teams translate emerging AI capabilities into practical visual development, production, provenance, and content workflows.
Talk to TTI