AI did not break Canadian film. It arrived at a moment when Canadian film and television were already held together by incentives, fragile labour…

AI did not break Canadian film. It arrived at a moment when Canadian film and television were already held together by incentives, fragile labour, gatekeepers, and bad math. That matters. A bad version of the creative-AI debate treats film workers as if they woke up one morning, saw a text-to-video demo, and panicked because the machine was shiny. That is not the industry the machine entered. Canadian screen work in 2026 sits inside a complicated stack: federal and provincial tax credits, production-services incentives designed to attract foreign shoots, Canadian-content rules meant to protect national culture, public funding bodies trying to stretch limited money, streamers retreating from the peak-TV spending cycle, a strike-disrupted North American production pipeline, and independent producers trying to make the arithmetic close when almost nobody gets rich from the actual film. So this chapter starts before the prompt. It starts with the production economy AI is landing inside. The central question is not "will AI replace artists?" That question is too blunt to be useful. The better question is: where does AI enter the workflow, who owns the underlying work, who consented, who gets paid, who loses leverage, and what kind of culture forms around the people trying to use or refuse the tools? You will leave with: a working map of the Canadian film-production landscape; the difference between production volume and Canadian creative ownership; the policy machinery of tax credits, Telefilm, the Canada Media Fund, and CRTC streaming contributions; the Strange Harvest and Erica Lapadat Janzen cases as local evidence of detection panic and legitimacy policing; the distinction between augmentation and substitution; the performer-likeness and digital-replica stakes; the certification question AI poses to Canada's CanCon points system; and a working test for evaluating any AI-in-film claim without collapsing into either boosterism or panic. ---

Canada as production infrastructure

Canada is a film country. It is also a filming location.

Those are not the same thing.

The first sentence is the cultural claim: Canadian writers, directors, actors, cinematographers, production designers, editors, VFX workers, producers, grips, gaffers, assistant directors, locations teams, makeup artists, sound workers, animators, and post-production crews make work with a Canadian sensibility, Canadian labour, Canadian places, and Canadian creative risk. The second sentence is the industrial claim: Canada is also one of the world's major production-service jurisdictions, a place where foreign producers can shoot efficiently with experienced crews, favourable exchange rates, deep infrastructure, and public incentives.

Put a number on the thing AI is entering. In 2024/25 the Canadian film and television sector generated roughly $10.2 billion in production volume, contributed close to $12 billion to GDP, and supported about 181,000 jobs — and that was the recovery year. Twelve months earlier, production had fallen 18.5% to $9.58 billion as the strikes and the streaming-commissioning slowdown bit (CMPA Profile 2024 and Profile 2025). This is not a boutique anxiety. It is a six-figure workforce, already knocked off its footing, watching a new cost-cutting tool walk onto the lot.

GOBLIN FACTS — put the workforce on the scale before you weigh the tool. Canadian film and television generated about $10.2 billion in production volume in 2024/25 and supported roughly 181,000 jobs, per the CMPA's Profile 2025 — and that was the rebound year, after production fell 18.5% to $9.58 billion the year before. AI is not arriving at a hobby. It's arriving at a six-figure workforce that already lost its footing once.

That dual identity sits inside the tax-credit architecture.

At the federal level, the Canadian Film or Video Production Tax Credit (CPTC) is the Canadian-content side of the machinery. Canadian Heritage describes it as a fully refundable tax credit, available at 25% of qualified labour expenditure, intended to encourage Canadian film and television programming and the development of an active domestic independent production sector. In plain English: if a production qualifies as Canadian under the rules, the federal system helps finance Canadian-owned work by rebating a portion of eligible labour costs.

The Film or Video Production Services Tax Credit (PSTC) is different. It is the service-production side. Canadian Heritage describes it as a tax credit at 16% of qualified Canadian labour expenditures for accredited productions, designed to enhance Canada as a location of choice for film and video productions employing Canadians and to secure investment. Eligibility can include a taxable Canadian corporation or a foreign-owned corporation. In plain English: the production does not need to be Canadian in authorship or ownership in the same way. The policy goal is to attract production spending and employment to Canada.

British Columbia's production-services credit makes the distinction even clearer. The Province of B.C. states directly that the production-services tax credit is available to both domestic and foreign producers and has no Canadian content requirement. B.C.'s basic production-services credit is listed at 36% of qualified B.C. labour expenditure for productions beginning principal photography after December 31, 2024, with additional regional, distant-location, digital-animation/VFX/post-production, and major-production credits layered on top.

🧌 GOBLIN CHECK — a maple leaf on the call sheet is not the same as Canadian authorship. A show can spend millions in Vancouver, hire Canadian crew, use Canadian tax credits, and still be foreign-owned IP made for a foreign buyer. That is not fake work. It is real work. But it is not the same thing as Canadians owning the story, the copyright, the backend, or the audience relationship. Production volume and creative sovereignty are cousins. They are not twins.

The service-production bargain has real benefits. It creates jobs. It builds crew depth. It sustains studios, vendors, gear houses, post facilities, VFX shops, and a thousand small businesses that do not survive on auteur cinema alone. It gives Canadian workers experience on large sets and complex shows. It makes Vancouver, Toronto, Montreal, Calgary, Winnipeg, and other production centres part of the global screen economy.

It also creates dependency. When Canada is serving foreign production, the greenlight often happens elsewhere. The IP is often owned elsewhere. The distribution decision happens elsewhere. The cancellation happens elsewhere. The residual structure, marketing spend, release strategy, and accounting logic often happen elsewhere. Canadian workers feel the boom, but they do not control the switch.

That is the first layer AI lands on: an industry with enormous Canadian skill and partial Canadian control.

---

Canadian content, Canadian ownership

Canadian-content policy exists because the market, left alone, will not reliably produce Canadian stories at scale.

That is the old broadcasting-policy logic, and it is still the logic underneath Telefilm Canada, the Canada Media Fund, CAVCO certification, provincial agencies, broadcaster obligations, and the CRTC's implementation of the Online Streaming Act. The mechanism changes over time; the basic problem does not. Canada sits beside the largest entertainment economy on Earth, in the same language market for most English-language work, with a smaller population and a smaller domestic capital base. Without intervention, Canadian screens fill with foreign work and Canadian producers become service providers to other people's slates.

The public system tries to push back. Telefilm funds feature-film development, production, post-production, marketing, festivals, and industry promotion. The Canada Media Fund supports television, digital media, documentary, experimental, French-language, Indigenous, and other streams of Canadian work. CAVCO certification determines whether a production qualifies as Canadian for tax-credit purposes. The CRTC regulates broadcasting contributions to Canadian programming.

The Online Streaming Act, which came into force in 2023, extended that logic to online undertakings. In its 2026 policy decision, the CRTC says the Act requires the Commission to modernize the Canadian broadcasting framework and ensure that online streaming services make meaningful contributions to Canadian and Indigenous content. The same decision sets a modernized Canadian programming expenditure framework, including contribution requirements for large Canadian broadcasting ownership groups and unaffiliated online broadcasting ownership groups, with attention to Canadian copyright ownership and production partnerships.

This is not nothing. It is a serious policy attempt to keep Canadian and Indigenous programming from being structurally drowned by global platforms.

But the gap between funding Canadian content and fostering enough Canadian-owned creative power is the uncomfortable part.

Tax credits lower the cost of production. Funding bodies help projects get made. Streamer-contribution rules can add money to the system. None of that automatically solves distribution. None of it automatically gives independent producers meaningful leverage against broadcasters, platforms, distributors, sales agents, financiers, and completion-bond requirements. None of it guarantees marketing. None of it guarantees that a Canadian film with Canadian authorship will reach a Canadian audience in a way that returns meaningful money to the people who took the risk.

This is the broken middle: Canada has a system for getting certain work financed, and a much weaker system for making sure that work finds an audience, recoups, and builds durable independent companies.

The result is a strange emotional economy. Canadian producers can be asked to carry national-cultural ambition, employment policy, diversity policy, regional policy, export ambition, festival prestige, and commercial viability all at once, while operating with financing structures that leave them undercapitalized and distribution structures that leave them weak. Everyone says they want Canadian stories. The business model often says: yes, but not enough to make the risk sane.

AI enters that world as both temptation and threat. If a tool can make a pitch deck cheaper, a proof-of-concept faster, a previs sequence clearer, or a temp VFX shot possible for a microbudget producer, the attraction is obvious. If the same tool is used by a better-capitalized company to reduce crew, weaken crafts, strip bargaining leverage, or appropriate style without compensation, the threat is equally obvious.

Two parts of the Canadian system push against the service-jurisdiction default, and both are worth naming before AI gets near them. The first is the Indigenous Screen Office (ISO), an independent national funder for First Nations, Inuit, and Métis screen creators built around the principle of narrative sovereignty — the idea that Indigenous stories should be controlled by Indigenous storytellers, not optioned and reshaped by others. The ISO launched in September 2021 with CA$40.1 million over three years, moved to permanent funding of roughly CA$13 million a year from 2024, now administers the Canada Media Fund's $10 million Indigenous Program, and receives mandated contributions under the streaming framework. This is the screen-content version of the argument Chapter 9 made about Indigenous data sovereignty: who controls the asset (the dataset there, the story here) determines whether a community is a subject of its own work or raw material for someone else's.

The second is Quebec. French-language films from Quebec have been Canada's top-grossing domestic films in six of the last seven years; Quebec sustains a genuine star system and a far higher domestic-market share than English Canada manages. It is the clearest Canadian counterexample to "we are mostly a service jurisdiction" — proof that a protected-language market with its own stars, its own press, and its own audience habit can keep its own stories on its own screens. But it is fragile, not finished: CBC and Globe and Mail reporting, alongside the broader record of Quebec cinema, noted French-market theatrical revenue falling about 45.5% in 2025. Even the strongest piece of Canadian creative sovereignty is one bad box-office year from looking like everyone else's.

The tool does not decide which one it is. The production economy does.

---

The certification question

Here is a Canadian question almost nobody in the AI-in-film debate has asked yet, and it sits directly on the machinery this chapter just described.

To certify as Canadian for the CPTC (to earn the 25% credit and qualify under Canadian-content rules), a production has to score at least six of ten points, and those points are awarded for the citizenship of the people in key creative roles: director, screenwriter, lead performers, director of photography, art director, music composer, picture editor. It also needs a Canadian producer. "Canadian," in the Income Tax Regulations, means a citizen or permanent resident. The CPTC and the service credits then rebate Canadian labour expenditure — money paid to Canadian people for their work (CAVCO / Canadian Heritage).

The entire test, in other words, is built on Canadian humans. AI is not a person. It holds no citizenship. It draws no salary. So what happens to the points, and to the labour-expenditure base, when generative tools do part of the writing, the boards, the design, the score, or the voice?

For now, the honest answer is not mentioned. The framework certifies the citizenship of people, not the provenance of machine output, and as of 2026 it carries no provision addressing generative AI. That silence cuts two ways, and both should worry someone. Read strictly, a production that hands a key creative function to AI could lose the points that role used to earn — quietly penalizing exactly the scrappy, under-capitalized Canadian work most tempted to use the tools. Read loosely, a production could keep its Canadian certificate while hollowing out the Canadian labour the credit exists to fund — collecting public money for "Canadian content" increasingly made by a model that is neither Canadian nor a worker. Either way, the instrument that defines Canadian screen culture is running on a definition written before the tool now reshaping how the work gets made.

🧌 GOBLIN CHECK — a points test for people, in an age of non-people. CanCon certification counts Canadian citizens in the key creative chairs. A generative model holds no passport and cashes no cheque. Until CAVCO says otherwise, "made in Canada" and "made by Canadians" can quietly drift apart inside the same certified production — while the credit keeps paying out. Watch which way the rules get read, and who benefits from the reading.

---

The pile-up before AI

The anxiety around AI in film is not only about AI. It is also about timing.

Canadian film and television entered the generative-AI moment after a pile-up of shocks. COVID shut down production, then restarted it under expensive health-and-safety protocols, insurance complications, schedule volatility, and burnout. The streaming boom created demand for more content, more stages, more crews, more post capacity, more everything — and then the end of peak streaming started pulling some of that demand back. The 2023 WGA and SAG-AFTRA strikes disrupted the North American production pipeline. The Canadian Media Producers Association's Profile 2024 report documented a major decline in total Canadian film and television production volume in 2023/24, with the impact concentrated in the disrupted production environment. Profile 2025 showed a partial rebound, but not a return to the old certainty.

That is what "AI panic" often misses. The floor had already moved.

For crews, the instability is practical. Fewer greenlights mean fewer days. Fewer days mean fewer hours. Fewer hours mean benefit thresholds and rent payments start getting interesting in the least fun possible sense. For independent producers, rising production costs mean the same grant or licence fee buys less production than it used to. For post and VFX workers, compressed schedules and global competition were already normal before generative tools entered the room. For writers and performers, streaming residual fights and digital-replica protections were already on the table before text-to-video models became good enough to scare anyone.

The strikes matter here because AI was not a side issue. The 2023 Hollywood labour disputes were about streaming economics, residuals, staffing, and working conditions, but AI and digital replication were part of the bargaining terrain. That has a direct Canadian echo. ACTRA's January 2025 Independent Production Agreement includes explicit AI provisions around synthesized performances, digital doubles, and machine learning. The Directors Guild of Canada has engaged Parliament on AI and copyright. Canadian screen workers are not watching a foreign fight from a distance; they are inside the same contractual weather system.

The end of the peak-streamer era matters too. When streamers were chasing subscriber growth, the logic was volume. More originals, more territories, more shows, more bets. When the logic shifted toward profitability, the industry rediscovered cancellation, write-downs, narrower slates, and harder scrutiny. AI enters just as executives are being told to reduce costs and increase efficiency. That does not make every AI tool a plot to fire people. It does mean workers are right to ask what incentive structure the tool is entering.

In a stable industry, a new tool can feel like an experiment. In an unstable industry, the same tool feels like a memo from the future saying the budget is smaller and the deadline did not move.

---

Indie producers and the broken middle

Independent filmmaking has always required a high tolerance for indignity. The AI era did not invent that; it just gives the indignity new software.

The basic indie problem is that film is expensive, collaborative, and risky. Even a small film requires people, gear, insurance, locations, post-production, legal paperwork, deliverables, music clearances, festival fees, marketing materials, and time. The costs arrive before the audience. The revenue, if it arrives, arrives late, through contracts that are difficult to read, difficult to audit, and often written by people whose kids will not be eating instant noodles if the film does not recoup.

Then comes distribution. A producer can finish a film, premiere at a festival, secure reviews, win awards, and still not reach a paying audience at meaningful scale. A distributor can acquire rights and spend little. A streamer can buy rights and bury the work in a catalogue. A sales agent can report expenses before revenue. "Hollywood accounting" is the name people give to the broader phenomenon: the project can look valuable in the culture and still show no profit on paper for the people without leverage.

Canada adds a specific version of the problem. Public funding and tax credits can make production possible, but they do not automatically create a healthy market for independent Canadian films. Broadcasters and streamers are gatekeepers. Festivals are gatekeepers. Agencies, managers, labs, funds, and commissioning editors are gatekeepers. Some gatekeeping is unavoidable; curation exists because attention is finite. But gatekeeping becomes a structural problem when access depends too heavily on networks, class position, geography, family connections, institutional fluency, or the ability to survive unpaid development time.

Nepotism does not always look like a villain twirling a moustache. Often it looks like someone "known to the room" getting the meeting because nobody has time to take a risk on someone outside the room. It looks like an emerging producer spending years learning how to speak the funding dialect while someone with the right surname already has a lawyer. It looks like a director being told to build an audience first, then being told the audience they built online is not the right kind of legitimacy. It looks like the same handful of names circulating through panels about disruption.

This matters for AI because the tool can cut both ways.

On one side, AI can lower some barriers. It can help a tiny team make a pitch image, a schedule, a budget draft, a lookbook, a rough previs, subtitles, temp music, or a grant draft. It can help a producer who does not have a studio development department approximate some of the machinery that better-capitalized companies take for granted. Used that way, AI can be a scrappy little lever against gatekeeping.

On the other side, AI can strengthen the gatekeepers. The same studio that already controls distribution can use AI to test concepts, compress development, imitate visual styles, and reduce paid exploratory work. The same platform that already controls audience access can demand more assets, more deliverables, more localization, more marketing variants, and more "efficiency" without paying proportionally more. The same producer who used to hire a junior artist for pitch work can decide the junior artist is now optional. The same funding body can be tempted to ask for more polished applications, because the tools exist, which quietly raises the floor for everyone.

That is the broken-middle lesson: a tool that lowers one cost can raise the expectations around every remaining cost. If the system is already tilted, efficiency does not automatically democratize it. Sometimes it just lets the tilted system move faster.

---

The Strange Harvest lesson

This book already disclosed the Strange Harvest story in the front matter because it is not decorative biography. It is a case study.

During the festival run of Strange Harvest, the cut included roughly 30 seconds of licensed AI-generated stills, used as placeholders, intentionally and disclosed. The team removed the material for theatrical release. The accusations did not stop. People kept identifying "AI" in sequences that were practical, or in images that were simply rough compositing, texture, compression, Photoshop work, or the normal weirdness of low-budget production under pressure.

The lesson is not "audiences are stupid." They are not. The lesson is more uncomfortable: detection confidence does not reliably track synthetic origin.

Once an AI label enters a media object, it can stick to the whole object. The label changes how people look. They start scanning for telltale signs. A hand looks strange. A background texture feels wrong. A face is too smooth. A shadow does something odd. In ordinary film language, those might be continuity errors, VFX constraints, compression artifacts, makeup choices, lens distortions, production compromises, or just taste. Under AI suspicion, they become evidence.

This cuts in both directions. Synthetic material can pass as real. Real material can be accused of being synthetic. Both are credibility harms.

For filmmakers, this is now part of the production environment. A disclosed AI use can be treated as proof of total contamination. A non-use can be disbelieved. A practical effect can be misread as generated. An artist can become trapped in a defensive posture, not explaining what they made but trying to prove a negative: no, that shot was not AI; no, that actor was real; no, that painting was not prompted; no, that background was not stolen from the machine's soup.

That is a miserable way to talk about art. It is also not a good way to do ethics.

Ethics asks what happened, who consented, what rights were involved, what harms were produced, what alternatives existed, and who benefited. Detection panic asks whether the image feels cursed. The first can produce policy. The second produces comment sections with a magnifying glass and a flamethrower.

---

The soft violence of AI art discourse

Erica Lapadat Janzen's BC + AI talk, "The Soft Violence in AI Art Discourse," gives this chapter its local artist-community counterpart to the Strange Harvest case.

The setup is small enough to be ordinary and large enough to matter. Erica, a Vancouver new media artist, made one AI-assisted art post on Threads. The post did not include a full process breakdown or a manifesto. The response became a dogpile: hundreds of replies, thousands of likes, roughly 850 replies collected for analysis. Instead of turning the thread into a glitch-art piece, she treated the replies as data.

She manually coded the replies into categories: identity and legitimacy attacks, skill or process policing, moral framing that was not structural, structural or platform analysis, support or affirmation, and ridicule or dismissal. Her reported finding was that more than 70% clustered around the non-structural categories: identity attacks, process policing, moral framing without structural analysis, and ridicule. In her reading, the thread was not primarily a discussion of AI ethics. It was social discipline.

That distinction is useful because it does not require dismissing genuine AI-art concerns. Training data consent is real. Creator compensation is real. Platform concentration is real. Style appropriation is real. The environmental footprint is real. Labour substitution is real. The book has spent eleven chapters building the machinery for taking those concerns seriously.

Erica's point is that much of the actual discourse was not doing that work. It was deciding who counts as an artist, what counts as craft, what counts as real labour, and which tools make a person illegitimate. The examples she named were not careful arguments about platform governance or copyright. They were insults and legitimacy signals: lazy, fake artist, and worse. AI was the trigger; artistic identity was the bruise.

That is the phrase worth carrying: soft violence. Not violence in the physical sense; not a claim that criticism is forbidden; not a shield against accountability. Soft violence here means informal collective punishment that presents itself as moral common sense. Nobody has to ban the artist. Nobody has to censor the artist. The discipline is social, continuous, plausible, and deniable.

The Q&A sharpened the case. The host noted that Erica's experiment involved training on her own work, not a grab bag of other artists' work. That did not stop the punishment. In other words, even the fact pattern that should have made the ethics conversation cleaner (artist-led experimentation using the artist's own corpus) did not prevent the thread from collapsing into person-policing.

This is where Erica's talk and Strange Harvest meet. In both cases, the public reaction was only partly about the specific AI use. It was also about credibility, belonging, and contamination. The AI label became a social object: once applied, it reorganized how people interpreted the artist, the work, and the artist's right to be there.

If we want ethical AI art discourse, this matters. A culture that cannot distinguish structural critique from personal punishment will not produce better norms. It will produce quieter artists, more defensive disclosure, and a lot of people who learn that the safest public position is certainty shouted at someone else.

---

Where AI enters the workflow

Most AI-in-film arguments go wrong because they talk about "AI" as one thing.

On an actual production, AI can enter at many points:

Development: brainstorming, research summaries, draft outlines, coverage, pitch materials, grant applications, lookbooks
Pre-production: scheduling drafts, budget scaffolds, concept art, storyboards, animatics, previs, shot references, location visualization
Production: script supervision tools, continuity references, transcription, translation, accessibility notes, camera tests, virtual production aids
Post-production: rotoscoping, cleanup, upscaling, noise reduction, temp VFX, rough comps, subtitles, dubbing, localization, music and sound experiments
Marketing and distribution: trailers, social clips, poster variants, audience analytics, metadata, press-kit assets, platform delivery materials

Those uses do not have the same stakes.

An AI-generated pitch image trained on a director's own material is not the same as a studio model trained on thousands of unlicensed artists. AI-assisted transcription is not the same as generating a synthetic actor. A producer using an LLM to organize grant notes is not the same as replacing a writer. A VFX worker using AI denoising to remove a repetitive technical burden is not the same as a company using automation to reduce crew without sharing the gains.

The workflow question is where the ethics starts.

The best use cases are usually assistive, bounded, disclosed, and reversible. They help a human do work the human still controls. They are not the only source of truth. They do not impersonate a real person. They do not claim rights they do not have. They do not hide the labour chain. They do not convert someone else's work into a private shortcut without permission.

The highest-risk use cases are usually substitutive, opaque, identity-bearing, and rights-ambiguous. They replace a craft, simulate a performer, imitate a living artist's style, generate final material without disclosure, or make it impossible for workers to know whether their own work is being used to train the machine that undercuts them.

EXAMPLE — the assistive end, and its catch. The genuinely useful end of this is not hypothetical, and it is worth naming so the chapter is not just a list of threats. AI transcription and captioning already do real work: CBC/Radio-Canada runs an AI pipeline that auto-transcribes well over a thousand files a day to support its subtitling staff, inside CRTC accuracy rules that still require human review for live broadcast. Localization is the bigger Canadian story. In October 2024 Inuktut, spoken by more than 39,000 Inuit, became the first Canadian Indigenous language on Google Translate, built with Inuit Tapiriit Kanatami and supporting both writing systems, with Google's own caveat that the tool "will still make many mistakes." The National Research Council's Indigenous Languages Technology project and Library and Archives Canada's use of AI handwriting recognition point the same way: assistive, bounded, supervised. But the benefit case has the same catch as everything else in this book. A model that helps revitalize a language can also harvest it without the community's consent. One Inuit technologist called that "basically a colonialism product," which is why Indigenous data-sovereignty principles (OCAP, Chapter 9) apply to the helpful uses too, not only the harmful ones.

This is why the augmentation-versus-substitution distinction matters, but only if we keep it concrete. "AI as a tool" is not an answer. A chainsaw is a tool. A pink slip is also a tool. The question is what the tool is doing in the labour relationship.

---

The sub-sectors: animation, VFX, and games

The feature film is where the AI debate gets loud, but it is not where most Canadians in this industry actually work. The dense employment is in the service sub-sectors: animation, visual effects, and games. That is also where the labour numbers are scariest right now, which makes it the easiest place in the whole chapter to blame the wrong thing.

Quebec is the cautionary tale. The province's animation and VFX sector shed two-thirds of its permanent jobs in two years, from 8,037 full-time-equivalent workers in 2022 to 2,603 in 2024 (Quebec Film and Television Council, reported by the Globe and Mail). It is tempting to file that under "AI took the jobs." The reporting does not support it. The collapse tracks a post-pandemic streaming bust, the cost shock of 2021–22 inflation and interest rates, the 2023 Hollywood strikes freezing greenlights, and, specific to Quebec, the province's March 2024 decision to cap its tax credit at 65% for international clients, which sent footloose work to cheaper jurisdictions. Generative AI is the new fear arriving on top of an already-bleeding sector, not the wound that opened it. Keeping that straight matters, because if you misdiagnose a tax-policy injury as an AI injury, you reach for the wrong fix.

Where AI genuinely is entering, the early Canadian uses look more like assistants than replacements, at least so far. Ubisoft Montréal's "Ghostwriter" drafts throwaway crowd dialogue for writers to choose from and rewrite (presented at GDC 2023, framed as assist-not-replace). Toronto's MARZ built AI de-aging and lip-sync tools used on major series. EA's Vancouver-area studios partnered with Stability AI on generating game textures. The industry's own survey reports studios adopting generative AI "primarily as a brainstorming and idea generation tool," with another quarter not yet using it but planning to (Entertainment Software Association of Canada).

That games sector is not small: $5.1 billion in GDP, about 34,010 full-time-equivalent jobs, and 821 active studios in 2023–24. And it carries the same ownership asymmetry as the film side: foreign-owned companies account for roughly 88% of total industry employment. Lots of Canadian work, less Canadian control. The federal Job Bank already rates "3D animation artist" a strong labour-surplus risk through 2033, for reasons that are only partly about AI.

The counterweight worth naming is the part of this economy AI does not hollow out: physical capital. Vancouver's Pixomondo and William F. White built what was, for a time, the world's largest LED virtual-production stage, the kind of infrastructure that keeps high-end work onshore. And the cost of the bust lands first on the people with the least cushion. Junior and emerging artists watch senior workers take entry-level jobs at a discount; below-the-line crews and workers outside the big three provinces feel it earliest. AI threatens the same bottom rungs (storyboard, layout, junior compositing) that the downturn already thinned. Whatever AI does to this sector, it does it to a workforce already standing on a smaller floor.

---

Augmentation versus substitution

The recent Hollywood examples are useful because they show the same distinction from different angles.

Martin Scorsese's reported work with Black Forest Labs was framed, according to trade-press coverage, around storyboarding and previsualization: using generative tools to communicate images, mood, and visual ideas to collaborators. The defence of that use is straightforward. Directors have always used tools to communicate what they see: sketches, references, photo boards, previs, animatics, mood reels, rough comps. If an AI tool helps a director express an image more clearly while the final work remains collaborative and properly staffed, the tool can plausibly be augmentative.

The critique is also straightforward. Storyboards, concept art, and visual development are not empty slots waiting for software. They are skilled crafts performed by people. A tool that helps a director bypass those people can be experienced by those crafts as displacement, even when the director experiences it as communication. Augmentation for the person with final authority can be substitution for the person whose craft used to mediate that authority.

James Cameron's publicly stated AI position makes the tension even clearer. As reported in his interviews, his pro-AI argument is mostly industrial: blockbuster VFX costs are too high, and AI may help artists work faster without simply firing half the staff. That is the best-case productivity argument. But Cameron has also, by his own account, drawn a line around AI-generated actors and performances, where the issue is not only efficiency but embodiment, consent, and the irreducible human presence of performance.

That distinction is useful. A tool that removes repetitive technical friction from VFX is one debate. A tool that simulates a performer is another. A tool that helps a director communicate to a department is one debate. A tool that replaces the department is another.

Paul Schrader's reported synthetic-star comments push the opposite direction: AI-native cinema as forecast, provocation, or bargaining threat. And the experiments are arriving. Trade reporting describes Hell Grind, a 95-minute AI-generated feature made with Higgsfield and ByteDance Seedance tools for a reported US$500,000 (mostly compute) in about two weeks, which screened in 2026 at Cannes' Marché du Film, the festival's sales market rather than the official selection, which bars wholesale AI films (Variety). The same reporting calls it visually convincing and narratively incoherent, which is the tell: feature-length generated media is becoming technically legible before it is aesthetically persuasive. The lesson is not that the all-AI feature has arrived as a mature form. The lesson is that the industrial imagination now has one more option on the table.

The director examples are about people with final authority. The studio scale is a different animal. In September 2024, Lionsgate signed a data deal with the generative-AI video company Runway, opening its library of more than 20,000 titles — franchises including John Wick, The Hunger Games, and Knives Out — to train a custom Runway model for storyboarding, previsualization, and VFX. Trade press (The Hollywood Reporter, Variety, Deadline) reported it as the first such partnership between Runway and a major studio. By 2026 Lionsgate had taken an equity stake in Runway and announced plans to draw on its catalogue for AI-generated short-form series. Hold that next to Scorsese. A director using a generative tool to communicate a shot is one thing; a studio licensing an entire catalogue of its workers' output as training data is another. The worker-side critique is that the boards, the comps, the design work, the performances baked into those 20,000 titles were made by people who were paid to make a film, not to feed a model that may later compete for their next job. And Runway, like the other major video generators, faces copyright litigation; those claims are pending and unproven, so file it as an open question rather than a verdict. The point is the category shift: when the training corpus is the company's own back catalogue, "augmentation" and "substitution" stop being a clean line and start being an org chart.

The chilling effect runs the other way too — toward the artists who decline. In late May 2026, Amazon MGM announced a GenAI Creators' Fund, with a flagship series sometimes referred to as "Project Nara." Within roughly two days, Jorge R. Gutiérrez (director of The Book of Life and Maya and the Three) publicly withdrew, after backlash from the creative community: "I have decided to drop out of the AI program at Amazon," as reported by Deadline (and corroborated by Variety and IndieWire). It is a small data point and a loud one. A filmmaker with real franchise credibility looked at the reputational math of being the face of a studio AI initiative and decided the cost was higher than the cheque. That calculation — not the technology — is what shapes who actually shows up to use these programs.

And once an option exists, someone will use it in a negotiation.

"We could do this with fewer people." "We could generate the boards." "We could use a synthetic voice." "We could make the proof of concept without hiring a crew." "We could localize without performers." Even when those claims are exaggerated, they shift bargaining power. A weak threat can still be a threat if the person hearing it cannot afford to call the bluff.

---

Performers, likeness, and consent

Film makes the likeness-rights issue concrete because performers are not abstract data subjects. Their bodies, voices, faces, timing, movement, and presence are the work.

ACTRA's position is therefore not a side note to the AI copyright debate. It is one of the clearest Canadian examples of AI becoming a labour and consent issue. ACTRA's 2023 member survey found overwhelming concern about misuse of name, image, and likeness rights and about AI replacing human actors in certain roles. The January 2025 ACTRA Independent Production Agreement includes explicit provisions against using performer recordings to simulate or alter a performer's voice or likeness, create synthesized performances or digital doubles, or train machine-learning systems without consent.

Those clauses matter because they answer the question at the level where workers actually live: the contract.

The law may be slow. Parliament may be slower. Copyright doctrine may be contested. But a collective agreement can say: you cannot use my recorded work to build a synthetic version of me without consent. That is not a complete solution. It does not protect every non-union performer, every background worker, every influencer, every ordinary person whose face is online, or every Canadian outside the contract. But it is a working mechanism.

It is worth seeing how concrete that mechanism has become, because the detail is where the protection lives. British Columbia's performer agreement now carries a dedicated AI section (Article F). Before a performer's "digital replica" can be made, the producer has to give at least 48 hours' notice, a "reasonably specific description" of the intended use, and obtain explicit consent that the performer signs or initials. If a replica is used in a scene in place of the performer, they are paid the minimum daily fee for each day they would have worked. And one clause does structural rather than individual work: a background performer's digital replica "shall not be used to meet the background counts," meaning a producer cannot fill a crowd scene with copies of the same person and call the extras hired (UBCP/ACTRA). Two cautions keep this honest. These clauses are new: most landed in 2024 and 2025, they are modelled on the U.S. SAG-AFTRA language and lag it by a year or two, and they have essentially no enforcement track record yet. Canadian performers spent that gap watching from the other side of the border; as one BC union member put it during an earlier contract extension that punted the issue, "for 18 months we're not going to talk about AI… the conversation will be over by the time we get there." The writers got further faster: the Writers Guild of Canada's 2024 production agreement requires producers to disclose AI-generated material, denies it writing credit, and bars training on members' work.

There is a Canadian voice-labour story the likeness debate usually skips, and it is one of the most AI-exposed corners of the whole screen economy: dubbing. Montréal is one of the world's major dubbing centres, and Quebec's doublage sector (the actors who give French voices to foreign films and series) sits squarely in the path of AI voice synthesis. The pressure is not hypothetical: YouTube rolled out optional multi-language AI dubbing in September 2025 after a two-year pilot, and Amazon Prime Video tested AI-generated dubs in March 2025. The Union des artistes (UDA) and the Association nationale des doubleurs professionnels (ANDP) represent these workers; UDA president Tania Kontoyanni has called AI "extremely concerning" for the sector and pressed governments to regulate it, and UDA has negotiated agreement language barring the use of members' vocal performances to train AI without consent (UDA; Quebec trade press). It is the same move ACTRA made on screen performance, applied to the voice — and it lands hardest in the one Canadian market, Quebec, whose linguistic distinctiveness was supposed to be its protection. A synthetic French track does not ask for residuals.

And here the clauses show their seams. The same BC agreement that demands consent before a performer's replica can be built carves out exactly this use: "When a production adjusts the voice to a foreign language, consent is currently not required" (UBCP/ACTRA). So the AI application pressing hardest on Quebec's doublage actors, swapping a performance into another language, is the one the new replica rules currently exempt. This is the enforceability ladder in miniature: a protection can be real, signed, and in force, and still have a hole cut precisely where the technology is moving fastest. Contract language is only as strong as its carve-outs are narrow.

SAG-AFTRA's 2023 and subsequent AI bargaining did similar work in the U.S. context, embedding digital-replica protections, consent rules, and compensation mechanisms into labour agreements. The crystallizing example, the thing that made digital-replica consent suddenly concrete for everyone, was the treatment of background performers. In the 2023 SAG-AFTRA dispute the union's account was blunt: a studio proposal to scan background actors and reuse their digital likenesses indefinitely for a single day's pay. "Scan once, reuse forever" is the union's characterization, and the studios disputed the specifics — but as a flashpoint it did its work, because it turned an abstract worry about "AI in film" into a picture anyone could hold: the lowest-paid, least-protected workers on a set being asked to sell their faces in perpetuity. The fight kept moving. The ratified 2026 TV/Theatrical agreement permits AI or synthetic performers only where there is "significant additional value" — language meant to cover both a working performer and a digital replica — and folds in a long-sought SAG/AFTRA pension-plan merger targeted for January 1, 2028. The numbers tell the rest of the story. Members ratified it with 91.4% in favour on roughly 19.3% turnout, with the studios adding about 1% (around US$38 million over the deal's final two years) and some on-record dissent at the board level (Variety; The Hollywood Reporter, Katie Kilkenny, May 12, 2026). The important point is not that the American settlement solves the problem. It does not. The important point is that the most advanced screen-labour response to AI is not a moral slogan. It is contract language.

EXAMPLE — the digital stand-in. On set, a stand-in holds the actor's spot while the crew lights the scene — a body, not a performance. AI "temp" work began as the same idea: rough it in now, replace it later. The whole fight on film sets right now is what happens when the stand-in turns out good enough to keep, and nobody asked the actor.

The DGC's parliamentary engagement sits beside this. Directors are not performers, but directors are also workers whose authorship, style, and creative decision-making can be affected by AI tools. Writers, editors, production designers, storyboard artists, cinematographers, composers, animators, VFX workers, and sound teams all have their own versions of the same question: when does a tool support the craft, and when does it appropriate the craft?

In June 2026 the DGC escalated from testimony to declaration, releasing a Manifesto on the Value of Human Creativity at the Banff World Media Festival. Two lines carry the weight: art is not content, and tools are not authors. The document frames AI's risk as cultural as much as economic, arguing that systems "optimized for scale" flatten distinct regional and national voices, and it asks governments, platforms, and funders for stronger human-authorship protections, safeguards against the displacement of creative workers, and standardized, mandatory disclosure of AI use. Read it as what it is: the position of a guild advocating for its members, not a neutral finding. But notice what kind of ask it is. Mandatory AI-use disclosure is exactly the mechanism this book keeps hunting for, and the DGC is asking for it precisely because it is not yet a binding requirement in Canadian law. A manifesto is a demand, not a duty. The guild paired it with a Low-Carbon GenAI Toolkit for estimating a production's generative-AI footprint, which ties the creative-labour fight back to the environmental one in Chapter 8.

The Beijing Treaty gap from Chapter 11 returns here. Canada has not signed or acceded to the Beijing Treaty on Audiovisual Performances, which means Canada lacks one international performer-rights commitment that would strengthen the moral-rights and economic-rights conversation around audiovisual performance. That is not an AI treaty. But AI makes the absence more visible.

Consent does not stop at the living, either. Synthetic performance makes it possible to revive a dead actor's face and voice for new work, turning "who agreed to this?" into a question about estates, heirs, and legacy — whose permission, on whose behalf, and for how long after the performer can no longer object. Canada has no likeness-rights statute built for that; here too, the contract clauses are doing the law's job.

The underlying principle is simple: a performer is not raw material.

ALIGNMENT — efficiency, or consent? When an AI tool gets pitched on a set, the case is almost always speed and cost. The question the pitch skips is whose face, voice, or craft the tool was built from, and whether that person agreed and got paid. Those are two different conversations, and the second one rarely makes the slide deck.

Neither is a director's style. Neither is a storyboard artist's line. Neither is a VFX worker's comp. Neither is a writer's voice. The harder legal question is how to operationalize that principle without freezing legitimate tools, fair dealing, parody, criticism, archival use, accessibility, and independent experimentation. The chapter does not solve that. It names the terrain.

---

The labour bargain

The film industry is not simply banning AI. It is trying to contractualize it.

That is a less dramatic sentence and a more accurate one.

The work ahead is not one universal rule. It is a set of bargains across many crafts:

Consent. Workers need to know when their work, likeness, voice, performance, writing, images, or style will be used in AI systems, and they need the right to refuse uses that change the nature of what they agreed to do.

Compensation. If a worker's labour creates reusable synthetic value, the compensation structure has to reflect that. A one-day session cannot quietly become a perpetual synthetic asset.

Disclosure. Audiences, workers, distributors, funders, and insurers need to know when AI materially affects a work, especially where identity-bearing or final-image material is involved. Disclosure cannot be so vague that it means nothing, or so punitive that honest disclosure becomes a confession booth.

Auditability. Producers and unions need records. What tool was used? What data went in? Was any performer likeness involved? Was any union-covered work used for training? Can the claim be checked later?

Scope. Not every use needs the same process. AI transcription and a synthetic actor are not the same category. A useful framework has to distinguish administrative support, development aids, temporary materials, final creative material, identity-bearing material, and training uses.

Worker participation. The people whose work is affected need seats at the table before the tools are normalized. A policy written by executives, lawyers, and vendors will miss the places where the craft actually lives.

This is where Canadian film could do something useful. Canada has a strong labour tradition, a dense guild and union ecosystem, public funding bodies, provincial agencies, and a cultural-policy system already used to attaching conditions to support. That ecosystem is wider than the performers and directors who get quoted most: it runs through the below-the-line technicians of IATSE, NABET, and the Teamsters, and the musicians of the Canadian Federation of Musicians (which signed its first Independent Production Agreement with the producers' association for 2025–2027) alongside the Screen Composers Guild of Canada, every one of them now facing its own version of the question, from AI-deskilled craft work to AI-generated scores. Those institutions can become frictionless enablers of substitution, or they can become the places where better norms get built.

The choice is not "AI or no AI." The choice is whether AI enters through contracts, consent, disclosure, and worker power — or through procurement, panic, and after-the-fact apology.

---

Provenance, detection, and audience trust

Film has always been artificial. That is not the problem.

Cinema is lenses, lights, cuts, makeup, sets, doubles, matte paintings, miniatures, CGI, ADR, colour grading, sound design, stunt work, fake blood, fake weather, fake rooms, fake dawns, fake wounds, fake cities, and actors pretending to be people they are not. The point was never literal reality. The point was trust in the terms of the artifice.

Generative AI changes those terms because it makes certain kinds of artifice cheaper, faster, more personal, and more identity-bearing. It also makes the record harder to inspect. A practical effect leaves a production trail. A VFX shot leaves files, vendors, contracts, plates, and invoices. A generated image may leave prompts, seeds, model names, metadata, and logs — or nothing visible at all.

That is why provenance matters.

OpenAI's provenance work, C2PA Content Credentials, SynthID watermarking, and similar systems matter because they try to preserve evidence about origin and manipulation. Bill C-34's synthetic-content labeling duties point in the same direction from the regulatory side. These systems are not magic. Metadata can be stripped. Watermarks can fail. Detection tools produce false positives and false negatives. But provenance is still better than vibes.

EXAMPLE — a receipt for a photo. Provenance tagging is just a receipt: this image came from this camera at this time, or this clip was generated by this model. It doesn't prove anything is true. It proves where the thing came from — which, in a flood of convincing fakes, is most of what you can actually check.

For film, the goal should not be a scarlet letter on every tool-assisted frame. The goal should be inspectability where inspectability matters: performer likeness, final-image synthetic material, documentary claims, news-adjacent media, marketing claims, and any case where the audience is being asked to believe a real person did or said something.

The Strange Harvest lesson is that audience detection cannot carry this burden. Erica's talk adds that social punishment cannot carry it either. If the only accountability mechanism is a crowd deciding what "looks AI," real artists will be falsely accused and synthetic deception will still get through. That is the worst of both worlds: panic without verification.

The better norm is boring and beautiful: keep records, disclose material uses, protect worker consent, distinguish final material from temporary material, and do not pretend the audience can solve provenance by squinting harder.

---

The working test

For any AI-in-film or AI-in-media claim, ask:

Ownership. Who owns the underlying IP, the training material, the generated asset, the final work, and the audience relationship?

Consent. Did the people whose work, likeness, voice, style, or performance is being used know and agree? Was the consent specific, revocable, compensated, and documented?

Labour. Is the tool supporting a worker, replacing a worker, deskilling a worker, or shifting work to lower-paid, less visible labour?

Disclosure. Is the AI use disclosed in a way that helps audiences, funders, unions, and collaborators understand what happened, or is it hidden behind "technology was used" fog?

Source. What model or tool was used? What data was it trained on? Is the claim "trained only on our own work" verifiable, or just a soothing sentence?

Distribution. Who gets paid if the work succeeds? Who has audit rights? Who carries the loss if it does not recoup?

Power. Does the tool give more leverage to independent creators and workers, or more leverage to the companies that already control financing, distribution, and audience access?

Critique. Is the objection structural (training data, consent, compensation, displacement, disclosure, platform power), or is it legitimacy policing dressed up as ethics?

That last question is not a trick. Structural critique is necessary. So is refusing soft violence. The fact that some people use ethics language as punishment does not make the ethics fake. It means the work has to be better.

---

What the set teaches

CHAPTER RECAP — you now have: - The Canadian film landscape separated into two overlapping realities: a real Canadian creative culture and a service-production machine that can employ Canadians without giving Canadians ownership of the story. - The policy machinery named: CPTC for Canadian productions, PSTC and provincial service credits for attracting production, Telefilm/CMF funding, and the CRTC/Online Streaming Act framework for streaming contributions and Canadian programming expenditure. - The pre-AI instability mapped: COVID, strike disruption, the end of peak-streaming expansion, rising indie production costs, distribution weakness, and gatekeeping as the conditions AI enters rather than causes. - Strange Harvest as a local detection-panic case: disclosed AI created a credibility label that kept attaching even after the AI material was removed. - Erica Lapadat Janzen's BC + AI "soft violence" analysis as a local artist-community case: roughly 850 replies coded into patterns where legitimacy policing often displaced structural critique. - The workflow map for AI in film, from pitch decks and previs to VFX cleanup, synthetic performers, marketing, and provenance. - The certification question: CanCon's points test rewards Canadian people, and has no answer yet for AI doing the creative work. - The service sub-sectors (animation, VFX, games) as the real employment core, where a two-thirds Quebec job collapse traces mostly to tax and market shocks rather than AI, and where the genuinely useful AI uses (captioning, Indigenous-language localization) still have to pass the consent test. - The concrete union mechanics: BC's Article F (48-hour notice, signed consent, replicas barred from filling background counts) and its foreign-language-dubbing carve-out, as a live example of a real protection with a hole cut where the tech moves fastest. - The working test for any AI-in-film claim: ownership, consent, labour, disclosure, source, distribution, power, critique.

The next chapter (Chapter 13) takes the synthetic-media and credibility problem into the public sphere: deepfakes, political misinformation, Bill C-16, synthetic-content regulation, and the empirical record from the 2025 Canadian election. Film taught us the audience cannot reliably detect what is synthetic by staring harder. Chapter 13 asks what happens when that problem moves from the set to the election feed.

You can now read any AI-in-media claim with the question the industry keeps trying to dodge: not "is AI good or bad," but who owns the work, who consented, who gets paid, and who is being asked to absorb the risk?

---

Bias label for this chapter: practitioner-informed structural analysis of Canadian film and media production as AI enters screen workflows. Author lean: written from inside the Canadian film community, sympathetic to workers, independent producers, artists, and performers facing AI-driven pressure; skeptical of both executive efficiency framings and anti-AI moral panics that punish individual creators instead of addressing platform, labour, copyright, and distribution structures. Government and regulator sources treated as primary on tax-credit and broadcasting frameworks. Industry association sources treated as useful but self-interested on production volume and economic impact. Union and guild sources treated as primary on worker positions. Personal and community case studies (Strange Harvest, Erica Lapadat Janzen's BC + AI talk) treated as lived evidence of discourse dynamics, not as comprehensive empirical measurement.

Primary sources cited or relied on in this chapter: Canadian Heritage, Canadian Film or Video Production Tax Credit (CPTC); Canadian Heritage, Film or Video Production Services Tax Credit (PSTC); Province of British Columbia, Production Services Tax Credit; Telefilm Canada and Canada Media Fund program documentation; Indigenous Screen Office (ISO) program documentation; CBC and Globe and Mail reporting on Quebec French-language box office and the Cinema of Quebec record; CRTC Broadcasting Regulatory Policy 2026-96 and related Online Streaming Act implementation materials; CMPA Profile 2024 and Profile 2025 production-volume summaries; ACTRA 2024 AI Submission Explainer and 2023 member survey; ACTRA Independent Production Agreement ratified January 21, 2025; Directors Guild of Canada parliamentary testimony on Bill C-27; SAG-AFTRA TV/Theatrical agreement materials, 2023 strike materials, and 2023 background-performer and digital-replica bargaining materials; BC + AI / Vancouver AI, Erica Lapadat Janzen, "Soft Violence in AI Art Discourse" (February 5, 2026 video page and transcript); OpenAI content provenance documentation; C2PA Content Credentials documentation; Google DeepMind SynthID documentation; and trade-press reporting on the film-AI shift: Variety on Martin Scorsese's advisory role with Black Forest Labs and the Art Directors Guild's response (June 2026), Deadline on James Cameron's VFX-cost remarks (Glenn Garner, April 9, 2025), Variety (October 2025) and The Hollywood Reporter (Katie Kilkenny, May 28, 2026) on Paul Schrader, Variety on the AI feature Hell Grind at Cannes' Marché du Film (May 2026), The Hollywood Reporter, Variety, and Deadline on the Lionsgate–Runway data deal and equity stake (2024 and 2026), Deadline on Jorge R. Gutiérrez's withdrawal from Amazon MGM's GenAI project (May 2026), and Variety and The Hollywood Reporter (Katie Kilkenny, May 12, 2026) on the 2026 SAG-AFTRA agreement's AI and pension terms; the Union des artistes (UDA) and Association nationale des doubleurs professionnels (ANDP) on Quebec dubbing and AI (with Quebec trade-press coverage); the Canadian Federation of Musicians' 2025–2027 Independent Production Agreement with the CMPA and the Screen Composers Guild of Canada; and reporting on YouTube's and Amazon's AI-dubbing rollouts; Quebec Film and Television Council animation/VFX employment figures (reported by the Globe and Mail); the Entertainment Software Association of Canada, Canada's Video Game Industry 2024; Ubisoft Montréal's "Ghostwriter" (GDC 2023) and Pixomondo/William F. White virtual-production reporting; Google and Inuit Tapiriit Kanatami on Inuktut in Google Translate (2024), the National Research Council's Indigenous Languages Technology project, and Library and Archives Canada on AI-assisted archives; and UBCP/ACTRA's Article F (BCMPA) AI provisions alongside the Writers Guild of Canada's 2024 Independent Production Agreement. Detailed citations in the Sources appendix.

---