AI-Generated Video Ads or a Real Shoot? How to Choose

Full disclosure: we sell both routes. Our AI video production team makes AI films for brands in Kerala, across India and in the UAE, and the same people plan real shoots and 3D product films. So this guide isn't a pitch for one method. It's the test we run on a brief before deciding what to generate, what to render and what to film.

You can hold our work to it. Our AI videos page plays our AI films in full, each labelled AI-made, among them a courier film for ZAJEL in Dubai and a resort film for Ynad Mount Resort in Wayanad. Prices and rules below were read on 4 October 2026 and carry their own dates.

What can AI video do well in 2026, and what can it not?#

AI video is now good at mood, motion and imagination. In films we've made, a helicopter flies a gift-loaded sofa over a city for a furniture retailer's Aadi sale, King Mahabali sits down on a new sofa beside the pookalam for Onam, and an empty white room furnishes itself piece by piece. None of those would have been sensible to shoot.

What it still does badly is exactness. In our work, AI drifts on logos, labels and product shapes, which is why we check those frame by frame and fix or cut anything that doesn't hold up. A model that has never handled your product can only make something like it, and that matters most when the ad's job is to show the thing people will actually receive.

The other limit is trust. Our working rule, as our AI videos page puts it, is that a real shoot still wins "when people need to trust what they see: your team, real customers, food, testimonials and finished spaces". The rules now point the same way. So the useful question is rarely "can AI make this?" It's "does the viewer need this to be true?"

When should you film for real?#

Film it when the ad asks the viewer to believe something about the real world. Three cases cover most briefs.

Your product. If the buyer is judging a fabric, a finish, a stitch, a dial or a portion size, they need the real thing, filmed or rendered from its actual design. A generated necklace, plate of biryani or sofa that is "close enough" sets up a disappointment at delivery. ASCI's new guideline also lists exaggerating product results or features through visuals among the ads that break its code even when they carry an AI label.

Your people. Founders, doctors, teachers, staff and customers belong on camera as themselves. ASCI bars using AI to fabricate a testimonial or endorsement that falsely suggests a real person used or recommended a product, label or no label. When a presenter carries the message, a real one earns trust faster: the Meta ads we made for VONA School of Design, a new design school in Calicut, put one presenter straight to camera, with the offer in big captions.

Your premises. Hotel rooms, showrooms, clinics, campuses and finished homes should be filmed as they are. ASCI's guideline prohibits "depicting non-existent locations in a manner that implies that they are real and they exist", and its example is a travel agency showing a hotel that doesn't exist or facilities it doesn't have. Our guides to real estate video marketing and hotel social media marketing cover how to film those places well.

For a simple presenter-led reel, filming is often the cheaper route too, as the cost section shows.

When do AI or 3D beat a shoot?#

Three kinds of brief favour rendered or generated video.

When the product must be exact but the camera can't go there. That's a job for 3D, which isn't the same thing as AI. For Bosq, an office chair brand from Kozhikode, we made three vertical product films in which each chair was built as a 3D model and rendered, not photographed. There was no set to build, the light could go exactly where the mesh reads best, and the camera could glide around and under each chair in a way a real one would need a rig for. Because the model is built from the product, the details stay true from every angle, which a generated chair can't promise.

When the idea lives in imagination, seasons or scale. Festivals, myths, weather, impossible stunts and journeys are where AI earns its place. Our film for ZAJEL, Every city has its rhythm, moves through sea and air freight, a sorting hall and Dubai's roads in 32 seconds, scenes that would need a port, an airport and a hall full of parcels if they were shot. ASCI's guideline even exempts "obvious, unrealistic effects that audiences recognise as not depicting reality" from labelling; its own example is a fairy repairing a product with magic dust.

When you need many versions, fast. Ramadan, Onam and Diwali creative, offers that change every week, language versions for Kerala and the Gulf and hook tests all reward a method that makes the tenth version almost as cheaply as the first. Our guide to short-form video formats that sell in the Gulf explains why Gulf ads wear out faster and need fresh cuts more often.

The decision table#

What the ad needs	Film it	Render it in 3D	Generate it with AI
Product truth: the exact logo, label, finish or shape	Yes	Yes, built from the real product	Risky: details drift, so check every frame
Real people: founder, staff, customers	Yes	No	Only with written consent and a label, and never as a testimonial
Your real premises	Yes	Only for spaces not yet built, labelled	No
Places, seasons or scale you can't reach	Costly	Possible	Strong
Many versions: languages, offers, hooks	Plan extra takes on the day	Re-render	Strong
Budget	Crew, kit, talent and location days	Modelling first, then cheap re-renders	Per-second tool fees, plus people to script, review and edit
Time	Slowest to line up: people, places and weather	Slow to model, quick to re-render	Quickest once the look is approved

Our planning view for a typical brief, not a price list or a quote.

What do AI video ads and real shoots cost?#

AI video has a visible price per second and a less visible one in people's time. The visible part, from Google's price list for its Veo 3.1 models, for video with sound:

Google model	720p	1080p	4K
Veo 3.1	$0.40 a second	$0.40 a second	$0.60 a second
Veo 3.1 Fast	$0.10 a second	$0.12 a second	$0.30 a second
Veo 3.1 Lite	$0.05 a second	$0.08 a second	Not offered

Paid tier, in US dollars, from Google's Gemini API pricing page (opens in a new tab) (last updated 1 October 2026, read 4 October 2026). Veo has no free tier, and Google charges only for videos that generate successfully.

The catch is that you pay for every take you generate, not only the ones you keep. A 30-second ad at 1080p is $12 of generation on Veo 3.1 if every shot works first time. If you keep one take in five, it's 150 seconds of generation: $60 on Veo 3.1, $18 on Fast or $12 on Lite (illustrative arithmetic, not our production data). Then come the costs no price list shows: the script, style frames, review, edit, voice, music and the licences that go with them.

Tools also disappear. OpenAI told developers on 24 March 2026 that its Sora 2 models and Videos API would be removed from its API, and they were shut down on 24 September 2026 (opens in a new tab). Anything built on them had to move to another model. Keep the script, style frames, prompts and edit files of every film, so the look can be remade on another tool.

A real shoot has no per-second fee, but it pays for crew, kit, talent, locations and editing, and its price follows the scope. One Kerala reference point: ReelsBuddy, the video production brand Marketlube runs, charges from ₹999 for a shoot-only package and from ₹1,999 for a full video (presenter, script, shoot and edit), ReelsBuddy's own prices, with 50+ presenters to choose from (its own figure; more on the ReelsBuddy case study). For a straightforward presenter video, that is why filming often wins on price as well as trust. We quote AI films and shoots per brief, because the price follows what the film has to show. For motion videos made from a software product's own screens, we publish a starting price on our SaaS explainer videos page, with ten examples of the format we made for Xale.

Which labels and disclosures are required?#

Four sets of rules apply to a brand running AI video in India, and they overlap.

India's IT Rules, since 20 February 2026. MeitY's amendment, notified on 10 February 2026 as G.S.R. 120(E) (opens in a new tab), defines "synthetically generated information" as audio, images or video made or altered by computer so that it appears real and could be taken for a real person or event. Routine editing, colour correction, noise reduction and accessibility work are excluded. The duties fall on platforms. A tool that makes such content must label it visibly, or with a spoken disclosure at the start of audio, embed permanent metadata where that is technically feasible and not let either be removed; large social platforms must ask users to declare synthetic content before it goes up, check the declaration and show a label (the rules as amended (opens in a new tab)). For a brand, that means answering the upload declaration honestly and never trying to remove a label.

ASCI's guideline, from late December 2026. ASCI released its Guidelines for Responsible Labelling of Synthetically Generated Content in Advertising (opens in a new tab) on 29 September 2026, and they come into effect three months from publication, according to ASCI's press release (opens in a new tab). They ask whether the synthetic content is central to the persuasion and whether leaving out a label would mislead. A label is mandatory for synthetic influencers, a replicated face or voice (even with consent), fabricated events or settings that change how the product is understood, products that don't exist yet (its example is a 3D model of an unbuilt housing complex) and sound effects central to the product. Minor edits, decorative backgrounds and music, obvious fantasy, text work and accessibility need no label. Suggested wording: "Audio/Video created using AI" or "Audio/Video enhanced using AI", or the platform's own label.

YouTube. Its disclosure policy (opens in a new tab) asks creators to disclose content that makes a real person appear to say or do something they didn't, alters footage of a real event or place, generates a realistic scene that didn't happen or creates music that is the video's main focus. Beauty filters, colour and lighting changes and clearly unrealistic scenes are exempt. YouTube labels some content itself: anything made with its own AI tools, anything carrying C2PA metadata and anything its systems detect. Creators who keep failing to disclose risk removal of content or suspension from the YouTube Partner Program.

Meta. Since 2024, Meta has required people to disclose organic posts (opens in a new tab) with photorealistic video or realistic-sounding audio that was digitally created or altered, and it shows "AI info" labels (opens in a new tab) on AI content. On ads, it labels creative made or significantly edited with its own generative tools, and an update on 1 June 2026 (opens in a new tab) said it would begin detecting ads made with third-party AI tools through industry-standard signals and label them in "About this ad". Social issue, election and political ads carry a separate disclosure duty for the advertiser.

Our rule is simpler. Every film on our AI videos page is labelled AI-made, and we follow each platform's disclosure rules wherever a film runs. We never recreate anyone's face or voice without their written consent. Our advice to any brand is one line: if a viewer could mistake a frame for a real person, place, event or product result, label it, whatever the minimum is.

Who owns an AI-made video?#

Contractually, the answer should be in writing before work starts. When we make a film, the client owns the final film we deliver; that is what our service page promises. Legally, the picture is less settled.

India. The Copyright Act names "the person who causes the work to be created" as the author of a computer-generated literary, dramatic, musical or artistic work, and the producer as the author of a film (Copyright Act, 1957, section 2(d) (opens in a new tab)). DPIIT's working paper on generative AI and copyright (opens in a new tab) (December 2025) dealt with training data and left the copyright status and authorship of AI outputs to a second part.
The United States. The US Copyright Office concluded on 29 January 2025 (opens in a new tab) that prompts alone don't make an output protectable, while human creative arrangement or modification can be, and that using AI as an aid doesn't bar protection.
Training data. The Delhi High Court refused ANI an interim injunction against OpenAI on 24 July 2026, and ANI's appeal is next listed for 5 December 2026, MediaNama reported (opens in a new tab) on 15 September 2026.

The practical answer: put the assignment in the contract, keep the scripts, prompts, style frames and edit files (they record the human choices), license music and fonts for the platforms and period you need, and remember that footage you film yourself is ordinary copyright. That makes hybrid films the easiest to protect.

Can you mix real footage and generated scenes?#

Yes, and for most brands the mix is the answer. Shoot what carries the claim, generate what carries the mood, and label the generated parts when they could pass for real.

The TasHeel Moral Academy account shows both halves. Its Malayalam story reels retell Islamic history over AI-generated painterly scenes, and Instagram shows its AI label on the most-liked of them; its presenter reels, which our team shot and edited, were filmed for real on a studio set. The stories travel, and the real presenter earns the trust.

A few rules keep a hybrid honest and watchable:

Film the product and the people. Generate the world around them only where it doesn't change what the product does or how it looks.
Grade everything together. Real and generated shots should share colour, light and grain, or the cut shows.
Check the product in every generated frame. Logos, labels and shapes are where models slip.
Label what could be mistaken for real. Renders of unbuilt projects above all, because ASCI names them.

For format ideas that suit each route, see our guide to corporate video ideas for Kerala businesses.

Frequently asked questions#

Are AI-generated video ads allowed in India?#

Yes, as long as the ad is lawful and honest: the rules are about labelling and misleading content, not about the tool. Since 20 February 2026 the IT Rules have required platforms to label realistic synthetic content, and from late December 2026 ASCI's guideline requires labels in specific cases and rules out fake testimonials and exaggerated results even when labelled.

Do AI video ads have to be labelled?#

If they show realistic synthetic people, places or events, expect so. Large platforms must collect a declaration and label such content under the IT Rules, YouTube and Meta have their own disclosure rules, and ASCI requires a label for synthetic influencers, replicated faces or voices and products that don't exist yet. Obvious fantasy doesn't need one.

How much does it cost to generate an AI video ad?#

On Google's Veo 3.1 models, generation costs $0.05 to $0.60 a second, depending on the model and resolution (Gemini API price list, last updated 1 October 2026). Because every take is charged, a 30-second ad usually means several times 30 seconds of generation, plus the people who script, review and edit it.

Can AI make a video of my actual product?#

It can work from photos and footage of your product, but generated details drift, so every frame needs checking against the real thing. When exactness matters, film the product or render it in 3D from its real design.

Is a 3D product film the same as an AI video?#

No. A 3D film is built from a model of the product and rendered, so its shape stays true from every angle. An AI video is generated from prompts and references and can change details from one frame to the next. The two can share an edit.

Get an honest call on your next ad#

Most briefs end up as a mix: the product and the people filmed, the imagined parts generated, and anything that could pass for real labelled. If you'd like a second opinion on yours, tell us what the ad has to show and we'll say what we would film, render or generate. You can also see how our video production team plans a shoot, or watch the films we've made with AI.

Changelog#

2026-10-10: Linked the VONA School of Design and ZAJEL case studies, and the SaaS explainer price now points to ten examples of the format we made for Xale.