How AI Is Changing Commercial Production (and What It Cannot Replace)

The most honest thing you can say about AI in commercial production is this: it is doing real work right now, mostly in places you would not put a camera anyway.

Google reported that advertisers used Gemini to generate nearly 70 million creative assets inside AI Max and Performance Max campaigns in Q4 2025 alone. That is not a prediction or a pilot program. That is the current state. But look at what those assets are: performance marketing copy, social media variants, banner ad iterations, the volume layer of the advertising business that nobody frames on the wall.

The Super Bowl spot still has a director. The automotive launch still has a DP. The brand anthem still has an editor making taste decisions on every cut. The premium end of commercial production has not been touched in any meaningful way. What has changed is everything below it, and it is worth understanding exactly what is happening and where the line sits.

What AI Is Actually Doing Right Now

The clearest wins for AI are in pre-production. Specifically in treatment development.

Before AI image generation, building a treatment meant sourcing reference photography from stock libraries, pulling film stills, licensing images, and spending days curating a visual narrative that approximated your intended look. A set of 12 style frames from a professional designer cost $3,000 to $5,000 and took three days. Midjourney can generate a comparable set of directional references in hours for the price of a monthly subscription.

That is a real productivity gain. Directors exploring twenty visual directions before committing to one is no longer a luxury of the biggest budgets. The iteration happens in pre-production where it belongs, not on set where it costs ten times as much.

Storyboarding has moved the same direction. AI boards are faster and cheaper than hand-drawn boards on low-budget projects and pitch scenarios. They are not better for complex sequences or specific blocking, where a skilled storyboard artist can draw exactly what the director envisions in ways that AI consistently misinterprets. But for early internal review and quick client alignment, they are genuinely useful.

In post-production, AI has hit almost every department.

DaVinci Resolve's Magic Mask v2 is cutting 40% off colorists' turnaround time on commercial projects. AI rotoscoping tools from Boris FX, Runway, and Resolve have turned a task that once required days of frame-by-frame manual work into a task that takes hours with clean results. Adobe Premiere's Generative Extend covers transitions and extends reactions using the Firefly Video Model. Auto-reframing from 16:9 to 9:16 and 1:1 is now batch-processable. Captioning in 27 languages happens automatically.

These are not dramatic. They are tedious tasks getting done faster. That is exactly what a tool is supposed to do.

On the music side, Suno is pulling $300 million in annual recurring revenue with two million paid subscribers generating complete songs from text prompts, including vocals and production. AI-generated voiceover through ElevenLabs works for rough cuts, temp tracks, internal presentations, and a growing slice of social and performance marketing content. Warner Music Group settled its lawsuit with Suno in November 2025 and struck a licensing deal, so legal music generation is a real and expanding option for work that does not need original composition.

The AI Video Tools, Accurately

The AI video generation story is more instructive for what it reveals about hype than for what it reveals about capability.

OpenAI's Sora was supposed to be the inflection point. It launched as a standalone app in September 2025, hit one million downloads in its first ten days, then dropped 32% month-over-month through December, dropped another 45% in January, and was shut down entirely on March 24, 2026. The stated reason: generation times of 3 to 8 minutes per 10-second clip, which were acceptable in late 2024, became untenable when competitors cut generation times by 60 to 80% while matching quality. More fundamentally, Sora never solved "high controllability, reproducibility, and mass production" for commercial use. The most hyped AI tool in production history lasted six months in the market.

What is actually working: Google Veo 3.1 delivers native 4K output with cinema-standard color science and generates up to 60 seconds of video with synchronized audio. Runway Gen-4.5 provides the best temporal consistency and motion control among premium tools, which is why Lionsgate used it for previz and VFX work and saved "millions." Kling 2.6 generates up to two minutes at 1080p and costs roughly 40% less per second than premium competitors, making it the volume option for social content.

What these tools produce reliably: generic B-roll, product shots with simple lighting, abstract and atmospheric content, backgrounds for compositing, social media content where production value expectations are lower. What they still cannot produce reliably: consistent human characters across multiple shots, specific emotional performances, accurate lip sync to dialogue, brand-compliant product representation, and anything involving hands or detailed physical interaction.

The February 2026 Svedka Super Bowl commercial was the first primarily AI-generated spot to run at the Super Bowl. It generated attention. It also generated backlash. Coca-Cola's AI holiday campaigns generated substantial criticism. The brands that have gone all-in on AI-generated final deliverables have largely experienced consumer rejection.

The quality gap between AI video and traditionally produced footage, at the close-up human performance level, remains significant and detectable to trained eyes. At the atmospheric background level, it is closing fast.

The SAG-AFTRA Provisions

The 2024 SAG-AFTRA agreements established the most consequential labor protections in this area, and any commercial director working under union contracts needs to understand them.

The core protection: studios cannot create or reuse a digital replica of a performer's voice or likeness without explicit, informed consent for each specific use. Digital replicas cannot replace background actors within coverage maximums. Producers must notify SAG-AFTRA when they plan to create and use synthetic performers.

The Narrativ agreement from August 2024 established a consent-based model for AI voice licensing in audio advertising, with performers setting their own price above SAG-AFTRA minimums on every job. Replica Studios has a separate SAG-AFTRA agreement specifically for AI voice usage.

What this means practically: if you are on a SAG-AFTRA production and want to use AI-generated voice, background talent, or digital replicas, there is a documented process. It requires consent. It requires compensation. It is not optional. The "AI rider" is becoming standard in talent contracts, and the time to understand it is before you are in a conversation with a talent rep on a job that is already in pre-production.

The legal landscape around AI-generated likenesses is moving fast. Over 45 states have enacted deepfake legislation. The TAKE IT DOWN ACT was signed federally in May 2025. Tennessee's ELVIS Act grants individuals property rights in their name, voice, and likeness. The EU AI Act requires disclosure of AI-generated content for campaigns targeting EU audiences regardless of where the work is produced.

Copyright ownership of AI-generated content remains unsettled. The U.S. Court of Appeals affirmed in March 2025 that the Copyright Act "requires all eligible work to be authored in the first instance by a human being." Purely AI-generated treatment imagery, style frames, and concept visuals likely have limited or no copyright protection. This matters in production contracts where usage and licensing are being defined.

What Still Requires a Human Director on Set

This is the part worth being direct about.

The jobs disappearing to AI are concentrated in volume production and technical roles: junior retouchers, technical 3D artists, computer graphic artists (down 33% in 2025), voice actors doing corporate narration and e-learning content, and junior editors handling selects and versioning. The advertising and PR sector lost 54,000 positions in a single year through May 2025. These are real losses.

What is not disappearing: the ability to walk onto a set and get a performance out of a person.

You cannot prompt your way to a take. You cannot generate the moment where a real athlete does something unexpected and you recognize it and call cut and build the spot around that moment instead of the one you planned. You cannot A/B test your way to brand judgment. You cannot replace the DP who reads the light shifting at 4pm and knows you have eleven minutes to get the shot and makes the call to move the whole setup forty yards left because the shadow line is better there.

The AI tools that exist are tools. They are useful in specific applications. They compress timelines and reduce costs in pre-production and post-production work that is mechanical. They do not replace the creative decisions that make a commercial effective, the directing, the performance elicitation, the color choices that serve brand identity rather than technical correctness, the edit rhythm that builds anticipation.

If your team is comparing a planning platform like HeroVideo.ai with a production partner, the practical question is whether you need more pre-production output or a finished campaign. GLM has a focused breakdown here: HeroVideo.ai alternative for finished video campaigns.

The directors who are doing well in 2026 use AI to explore more options faster in pre-production, without letting those tools set the creative vision. They understand AI capabilities well enough to incorporate them meaningfully in treatments. They stay current on what the tools can do without chasing every new release. And they know the difference between AI as a production tool and AI as a creative substitute.

The market data supports this. Premium brand advertising budgets have not decreased despite AI availability. Super Bowl ad spending hit record highs in 2026 at $8 million or more for a 30-second spot. The brands paying those rates are still calling a director, a DP, and a crew.

The volume end of the business is being restructured around AI. The premium end is not. Understanding which part of the market you are in, and which part you want to be in, is the actual strategic question for commercial production professionals right now.

---

The honest version of the AI conversation in production is not about whether the machines are coming for your job. It is about which jobs are structurally vulnerable and which are not, and why. Performance marketing content is vulnerable. A brand film that has to do something true about a product is not. The difference is whether the work requires human judgment and taste or whether it requires throughput and variation at scale. AI is very good at the second thing. It has not gotten close to the first.