AI Commercials on TV: How They’re Made and What They Cost

In June 2025, a fully AI-generated commercial for the prediction market Kalshi ran in the YouTube TV stream of Game 3 of the NBA Finals. One filmmaker made it in about two days with Google Veo 3, and Kalshi said the AI prompting cost was under $2,000 (NPR). That story made AI commercials look like a $2,000 shortcut to TV.

Brand work looks different. A real product has to appear exactly as it does on the shelf, a face has to hold up on a 65-inch screen, and the spot has to pass a station’s technical check. Below is how we make AI commercials for TV and connected TV (CTV), where the technology still breaks, and what three of our recent AI ads actually cost.

What counts as an AI commercial

An AI commercial is a video ad in which some or all of the footage is generated by AI models instead of being filmed. In practice there are three levels:

• AI-assisted. The spot is filmed or animated the usual way, and AI helps with the script, storyboard, voiceover or cleanup.
• Hybrid. AI generates the scenes, people and locations, while the product, logo and interface are built in 3D and composited on top.
• Fully generated. Every frame comes out of an AI model, as in the Kalshi spot.

For TV ads with a real product, hybrid wins almost every time. AI models still distort packaging, invent letters in logos and shift brand colors from shot to shot. That is fine for an absurd concept spot and unacceptable for a snack brand whose bag has to match the store shelf. Most AI generated advertising that reaches TV today is hybrid, even when it is marketed as “made with AI”.

Why AI ads are moving to TV and CTV

AI in TV advertising is the newest chapter. Generative AI applications in consumer marketing started with social: quick variants, UGC-style ads and product clips. TV is the next step, and connected TV is what makes it possible for mid-size brands.

CTV inventory on platforms like Hulu, Roku and YouTube TV is bought on a CPM basis, with targeting by region and audience, so a brand does not need a national buy to get on the big screen. Our breakdown of what TV and CTV airtime costs in 2026 covers the media side. What changes with AI is the production side: CTV campaigns run several cutdowns and variants, and AI makes 5-, 10- and 15-second versions far cheaper than a reshoot. If you plan to run the spot on streaming, our connected TV advertising team handles both the creative and the placement.

How an AI TV commercial is made, step by step

Here is the AI commercial production process we used for RIVET Protein Puffs, a US protein snack brand. The brief: a brand awareness spot with no film shoot, a 10-second hero cut and a 5-second cutdown for local TV and CTV.

1. Write the script around what AI can’t do

The first decision was technical. AI still animates mouths and chewing badly: the face glitches and the snack breaks apart mid-bite. So we cut the eating shot from the script. The spot shows the craving, not the chewing: the product in the hero’s hand, the family waiting by the grill. The 10 seconds break down into a 2-second wide shot, a 3-second close-up, a 2-second medium shot and a 3-second packshot.

2. Lock the look without AI

Before any prompt, we built a moodboard from reference photos, palettes and locations. Then we showed the client several AI-generated casting options for the family. Agreeing on “active outdoor lifestyle” on a moodboard takes an hour; discovering a mismatch in the final edit takes a week.

3. Generate stills, then motion

Every shot is first approved as a still frame, then animated. Shots are planned for the edit from the start: a wide establishing shot, then closer shots, so the clips actually cut together.

4. Fix what the model gets wrong

This is where most of the hours go. Three characters in one frame tripled the usual problems. Hands glitched in most first passes, so we froze the one clean phase of the son’s hand movement and let the body move underneath. The father’s eye shifted by a few pixels, enough for viewers to feel that something was off. The grill turned about 90 degrees between shots, a second grill appeared out of nowhere, and by the tenth generation its metal texture started to melt. Each fix was done by hand.

5. Build the product in 3D

The RIVET bag never went through an AI model. We rendered it in 3D and composited it into the generated scenes, so the logo, typography and colors match the brand book in every frame. When the bag color drifted between two scenes with different lighting, we re-rendered both against one color reference.

6. Sound

A 10-second voiceover is a compressed copywriting brief: every word has to land on the picture. On top of the voice and music we added outdoor ambience and subtle effects, then balanced loudness for each placement.

7. Deliver for broadcast

AI clips come out at web resolution, and a TV screen shows every flaw. The final step is upscaling, resizing for each placement, loudness compliance and the station’s or platform’s technical specs. RIVET took 5 weeks from brief to delivery. For spots that need real actors and a crew, see our live-action TV commercial production.

Three AI commercials we made and what they cost

Here is what each project was, how long it took and what it cost:

• RIVET Protein Puffs: a 10-second spot in two edits plus a 5-second cutdown, hybrid AI and 3D. 5 weeks, about $6,000–7,000. Aired on local TV in one US state and on CTV.
• AMPD: a ~30-second ad, AI scenes plus 3D motion graphics. 4 weeks, about $5,500. Runs on the web, on the client’s homepage.
• SNEI: a 90-second animated film in a Pixar-like style. 4 weeks, budget not disclosed. Runs on the web.

RIVET: the TV spot

The only one of the three made for broadcast. The budget went mostly into the fixes described above and the 3D product layer, not into generation itself. No film crew, no location rental, no casting, and the brand still got a pixel-accurate package in every frame.

AMPD: two worlds in 30 seconds

AMPD needed to contrast a B2B marketer in an office with a shopper at home. Stock footage couldn’t match the locations and looks they wanted, so both worlds were generated with AI. The app interface on the phone screen was deliberately not generated: prompting AI to draw a UI inside a phone produces endless artifacts, so we composited it on top, along with a custom 3D scene. The spot went from brief to delivery in 4 weeks and now sits in the hero section of the client’s homepage. The same pipeline is ready for CTV; it simply wasn’t bought for TV.

SNEI: character animation without a 3D pipeline

SNEI, an environmental consulting firm in the Mojave Desert, wanted a story told by the animals it protects: a tortoise, a kit fox and a burrowing owl. Traditional 3D character animation at this length needs a large team and a six-figure budget. We generated the stills, animated them with Google Veo 3 and cast AI voices, but the accuracy came from people: the first fox came out as a generic red fox, and the tortoise was the size of a biologist until we fixed the scale.

What drives the cost of an AI commercial

Our hybrid brand spots landed at $5,500–7,000. A fully generated 30-second concept spot with no real product would come in at around $6,500 by our estimate: there is no 3D layer, but 30 seconds of footage takes far more generations and fixes than 10. Headline figures like Kalshi’s “under $2,000” cover the prompting only; the filmmaker’s fee was not disclosed. Most of the budget goes into the work around the generation:

• Characters in the frame. Every extra person multiplies the hand, face and eye fixes.
• Product accuracy. A real package, logo or app screen means a separate 3D or compositing layer.
• Iterations. A shot can take dozens of generations before one is usable; Kalshi’s 30 seconds took 300–400 generations for 15 clips.
• Versions. Each extra length, aspect ratio or language adds edit and QC time, though far less than a reshoot.
• Broadcast delivery. Upscaling, loudness and station specs are fixed costs that web-only ads skip.

For comparison, a traditional shoot adds crew, locations, talent, travel and insurance; our breakdown of why traditional video ads cost so much walks through those line items.

Where AI still breaks on a TV screen

A phone screen forgives a lot. A living-room TV does not. The weak spots we plan around on every project:

• Mouths and eating. Chewing, talking close-ups and lip-sync still slip into the uncanny valley.
• Hands. Fingers merge, multiply or lose their shape on contact with objects.
• Space between shots. Objects move, rotate or duplicate from one shot to the next, as RIVET’s grill did.
• Logos and text. Models hallucinate letters and drift colors, which is why the product goes in as 3D.
• Busy backgrounds. Crowds and streets show flicker and morphing at the edges of the frame.

Our vlog episode on the real limitations of AI video generation goes through these failures shot by shot. The other risk is the audience: some AI holiday ads from big brands drew public criticism, as we covered in 7 examples of AI-generated video ads. The fix is the same as with any ad: a real idea and a clean frame.

The producer’s view: clients usually expect less than they get

The most common surprise on our AI projects runs in the opposite direction from what people expect. Clients arrive with lower expectations, because the AI ads they have seen were made two or three months earlier, and in this field that is a long time. On a recent fintech commercial where we built AI avatars of the client’s own team, the reference work we showed in the pitch was already weaker than what we could deliver by the time production started. New models and tools come out every few weeks, so almost every AI project we finish is better than the same project would have been a month or two before. In practice the client nearly always gets more than they expected.

When to choose AI and when to shoot

AI is the right call when:

• the idea needs locations or scale you can’t afford to film;
• you need many versions and cutdowns for CTV and social;
• the timeline is weeks, not months;
• the product can be shown as a 3D render or a packshot.

Shoot it when:

• real people speak to camera in close-up, especially founders, doctors or ambassadors;
• the product has to be used on screen: eaten, poured, applied;
• trust is the message and any hint of artificiality would hurt it.

Many of our spots mix both: real footage where trust matters and AI to extend it, as in our Adidas AI commercial with David Beckham. For a fuller list of situations where AI pays off, see the vlog episode AI Video Tools Are Useless?

FAQ

Can AI commercials air on TV?
Yes. AI-generated ads already run on local TV and streaming platforms, including Kalshi’s spot during the 2025 NBA Finals and our RIVET spot on local TV. The spot has to pass the same technical and content checks as any other commercial.

How much does an AI commercial cost?
Our recent AI commercials cost roughly $5,500–7,000, and a fully generated 30-second concept spot comes in around $6,500 by our estimate. The price depends on characters, product accuracy, versions and broadcast delivery, not on the generation itself. Low figures in the news usually cover generation only, not the creative team.

How long does it take to make an AI TV commercial?
Our three projects took 4 to 5 weeks from brief to delivery, including approvals. A simple concept spot can be done in days.

Do viewers notice AI ads?
They notice bad ones: glitchy hands, odd faces, warped logos. A hybrid spot with a 3D product and hand-fixed frames reads as a normal commercial.

Can AI keep my product and logo accurate?
Not on its own. We build the product in 3D and composite it into AI scenes, so the package matches your brand book in every frame.

Planning an AI commercial for TV or CTV?

Tell us about your product, your audience and where the spot will run, and we’ll come back with a concept, a timeline and a budget. See our AI commercial production for brands or request a quote.