Ask Skip: Where Does AI Video Production Actually Make Sense?

Do you think AI video production has a meaningful role here in the future? If so, where does it make the most sense?

— Anonymous

Yes. In some parts of the production pipeline, a lot.

“AI video production” has become a terrible umbrella term. It lumps together previsualization, storyboards, background generation, VFX cleanup, localization, versioning, promo assets, digital replicas, synthetic performers, and fully generated scenes as if they’re all the same thing. They carry very different economics, creative value, and risk.

The cleanest rule is simple: AI makes the most sense when the human creative decision has already been made and execution is the expensive, repetitive part. The closer the tool moves toward performance, authorship, and the thing audiences are actually paying to feel, the more careful everyone needs to get.

Put AI Behind the Camera First

The easiest AI use cases are the ones most viewers will never notice.

Previs. Storyboards. Look development. Concept art. Background plates. Set extensions. VFX cleanup. Missing frames. Temp effects. Rough animation. Localization. Reformatting. Asset versioning.

That’s already where professional tooling is heading. Adobe is building generative video into storyboarding, concept visualization, B-roll creation, editing, shot extension, and other existing creative workflows. Netflix’s production guidance already assumes GenAI will show up inside real productions and requires partners to flag intended uses so talent, intellectual property, provenance, and creative risks can be evaluated before somebody creates a very expensive problem.

Netflix also spent $587 million to acquire InterPositive, Ben Affleck’s filmmaking technology company, putting production-focused AI capabilities directly inside the company. The technology is designed around problems such as visual consistency, lighting, missing shots, and post-production rather than generating a finished movie from a prompt. That’s a much more useful signal than another demo of a dragon wearing sunglasses.

The InterPositive acquisition pushes Netflix further into the production pipeline because the valuable part of AI may be eliminating friction between creative decisions, not automating the creative decision itself.

Preproduction Is Where AI Earns Its Keep Fastest

Preproduction is basically a giant exercise in spending small amounts of money to avoid spending huge amounts of money incorrectly.

A director can test visual approaches before building a set. A production designer can explore environments before committing resources. A VFX supervisor can communicate what a sequence might require. A producer can show financiers something closer to the intended result. Teams can kill weak ideas earlier and spend real production money on the ideas that survived.

Traditional production makes experimentation expensive surprisingly quickly. Every additional location, setup, concept, asset, revision, or handoff starts consuming labor and time. AI can make early experimentation cheap enough that teams try more things before the meter really starts running.

That’s the same problem sitting underneath the broader AI cost-cutting debate. Using AI to remove people from a spreadsheet produces an immediate number for the CFO. Using it to let a director test five approaches before choosing one is harder to quantify, but that second use can actually improve what ends up on screen.

Versioning Is the Obvious Volume Business

Streaming multiplied the number of downstream assets required for each title.

One piece of programming can require trailers, teasers, cutdowns, social clips, thumbnails, multiple aspect ratios, localized versions, promotional graphics, ad creative, metadata assets, and different packages for different markets and distribution environments.

Humans still need to decide what the campaign should communicate. Producing every variation manually becomes harder to justify as generation and editing tools improve.

AI is extremely well suited to this layer because the creative boundaries are already defined. The title exists. The characters exist. The brand exists. The campaign strategy exists. The job is producing more usable executions without requiring the same labor for every variation.

That’s where the economics described in infinite execution across media workflows become very real. When execution gets cheaper, media companies can create more versions, localize more aggressively, experiment with more formats, and support smaller audience segments that previously couldn’t justify bespoke creative work.

That will hit marketing, localization, and promotional production faster than prestige filmmaking because the volume is higher and the downside of any individual asset failing is lower.

Mid-Budget Production Has a Legitimate AI Problem to Solve

Film and TV production already has a cost problem. Shoots move toward cheaper jurisdictions, VFX budgets get squeezed, crews face shorter schedules, and projects get killed because the financial gap between ambition and expected return keeps getting harder to close.

AI can widen what’s financially possible without requiring every project to look smaller.

Amazon MGM has used AI alongside live-action footage on House of David to expand battle sequences, while its broader AI Studio initiative has focused on character consistency and integration with existing pre- and post-production tools. Lionsgate has similarly positioned its Runway relationship around pre-production, post-production, and more capital-efficient creation.

Amazon is now going further by building generative AI into a studio production system rather than treating it as another standalone creative app.

A filmmaker who can afford a more ambitious sequence because AI reduces certain VFX costs has gained creative capacity. A producer who can keep a project alive because virtual environments reduce location requirements has gained financing flexibility. An indie creator who can demonstrate an idea convincingly enough to attract distribution has gained bargaining power.

Reducing the cost of executing a shot preserves the creative decision with the filmmaker. Automating that decision moves AI into authorship.

Synthetic Performance Is Where the Trouble Gets Expensive

Once AI touches a recognizable face, voice, body, or performance, the economics change immediately.

That person represents labor, intellectual property, contractual rights, audience expectations, and often a meaningful piece of the project’s commercial value.

SAG-AFTRA’s 2026 TV/Theatrical agreement expanded protections around digital replicas and synthetic performers after earlier agreements established consent and compensation requirements. The use case still exists, but the cost calculation now has to include permission, bargaining, compensation, contractual restrictions, and reputational exposure.

A model removing a wire, extending a set, fixing continuity, or generating a background presents one category of risk. A model replacing a performance presents another.

Studios will explore both, but synthetic performance brings a much heavier mix of audience, talent, guild, contractual, and legal exposure.

They’d also be crazy to assume audiences, talent, guilds, and courts will treat every application equally.

Fully Generated Video Still Has a Judgment Problem

AI-native films are getting dramatically more coherent.

Recent projects can maintain characters longer, produce more convincing camera movement, create usable environments, and hold together across scenes in ways that would’ve looked impossible a short time ago. Some hybrid productions are already combining human actors with AI-generated environments and effects, while fully synthetic productions are testing whether recognizable talent can be licensed into generated performances.

Technical quality is improving quickly. That raises the value of the things the model doesn’t automatically solve: story, performance direction, pacing, chemistry, restraint, and judgment.

A production can generate an endless number of shots and still have no idea which shot belongs in the movie.

People keep asking whether AI can create something that looks expensive. It increasingly can. The business question is whether cheaper execution produces something people actually want to spend time watching.

The Labor Question Is Also a Product Question

Some production work will disappear. Other jobs will change. New specialties will emerge around models, pipelines, asset control, provenance, rights management, and AI-assisted creative workflows.

Pretending every existing job survives untouched insults everyone involved.

The management decision is what happens to the productivity gain.

A studio can use AI to make the same slate with fewer people. It can use AI to make more projects with roughly the same resources. It can increase production value on projects that previously couldn’t afford it. It can reinvest savings into development, talent, marketing, or experimentation.

The stronger economics come from removing rework, compressing handoffs, testing ideas earlier, increasing the number of viable creative options, and spending human time on decisions the audience can actually feel.

AI can reduce the cost of production. It can’t manufacture audience demand for something people didn’t want in the first place.

AI Makes the Most Sense Where Failure Is Cheap to Correct

Previs, post-production, VFX augmentation, localization, versioning, marketing assets, background generation, rough animation, restoration, and low-risk creative experimentation are the obvious places to push aggressively.

Fully generated narrative and synthetic performance require a higher bar. Rights need to be clean. Talent needs to be protected. Creative leadership needs to know why AI improves the project rather than simply making production cheaper.

AI video earns its place when it gives creators more options per dollar and more control over what ends up on screen.

Ask Me Anything

Whether you’re fed up, fired up, or just want the truth behind the trends, send me your questions using this form. Anonymity guaranteed. Bullshit not included.