In performance advertising, a creative doesn't last long. A video that brings in cheap leads at launch wears out with the audience after a while, and the cost per result climbs. You need a new video, then another one. If each of them requires a shoot day, actors or an influencer, the team runs out of production capacity before it runs out of ad budget.
AI video ads solve exactly this problem: volume. They don't replace a brand shoot and don't guarantee results, but they let you test more hypotheses faster. Here is how production of these videos works, why they need so many variants, who the format suits and which risks to account for up front.
What AI videos are and how they differ from a shoot
In performance marketing, an AI video usually means a short vertical video of 15–30 seconds in which the key elements are created by AI: the on-screen presenter (an AI avatar), the voice, part of the footage, sometimes the background and scenes. Real shots of the product, interface or packaging can and should still be used, since they add credibility.
The difference from a traditional shoot is not "image quality" but the economics of production.
| Filmed video | AI video | |
|---|---|---|
| Production | Shoot day, location, lighting, logistics | Generation and editing |
| A new variant | Reshoot | Edit the script and regenerate |
| Presenter | Tied to the shoot day | Available anytime, sticks exactly to the script |
| Versions for platforms, segments and offers | Each one needs production | Easy to make |
| Complex actions, tactility, natural facial expression | Stronger | Still weaker |
In exchange you get limitations: AI video still struggles with complex actions, tactility and natural facial expressions in long takes. That's why AI-generated ad video works best in formats like a talking head, problem and solution, screen demos and short skits.
How AI video production works
Take away the magic and the process looks like regular production, with the shoot stages replaced by generation:
- Product and audience. Who we sell to, what their pain points and objections are, which offers exist, what has already been tried in advertising and with what results.
- An AI brand avatar. The character who will appear in the videos: generated from scratch or based on a real person with their written consent.
- Voice. A synthetic or cloned voice, again only with the owner's consent.
- Content strategy and plan. Which formats, topics and hypotheses we test in the next cycle.
- Scripts and storyboards. The first two or three seconds (the hook) are written in several versions, because they decide whether people keep watching.
- Generation and editing. Avatar, voiceover, footage, subtitles, product inserts, 9:16 format.
- Review. Generation artifacts, lip sync, accuracy of product claims, legal requirements for advertising.
On the business side, you need very little to start:
- a description of the product and offers;
- an understanding of the target audience;
- your logo and brand colors;
- real photos or videos of the product and interface, if you have them;
- access to the results of past ad campaigns.
The more honestly you share what hasn't worked, the fewer hypotheses get wasted.
The key difference from a one-off shoot is that this is a cycle. After launch the results are reviewed, and the next batch of videos builds on what worked.
Why so many creatives: A/B tests and ad fatigue
Ad platform algorithms find the audience for a creative on their own, and much of what's required from the advertiser is to give them enough different videos to test. The more good variants you have, the better your chance of finding the one that delivers results more cheaply than the rest.
What usually changes between variants:
- The hook. A question, a provocation, an unexpected shot, a number. The most influential variable.
- The offer. A discount, a free trial, delivery, a guarantee.
- The angle. The customer's pain point, a comparison with an alternative, a testimonial, a how-to.
- Presenter and voice. Different avatars for different segments.
- The call to action and the final frame.
Without end-to-end analytics, test results are easy to misread: goals and events have to be set up before the ads go live.
Ad creatives wear out at different rates depending on audience size and budget. That's why a steady stream of new videos matters more than one "perfect" one.
A steady stream of new videos matters more than one "perfect" one.
UGC videos without influencers: possibilities and limits
The UGC format is a video that looks like content from an ordinary user: a person on screen talks about the product, shows how they use it, shares their impression. These videos blend into the feed and often outperform polished ads.
An AI avatar lets you make UGC videos without influencers: no searching, negotiating or waiting, and no risk that the creator disappears or films the wrong thing. But honesty matters here.
Who AI videos suit, and who they don't
A good fit
- online stores and e-commerce brands with a wide range and constant advertising;
- mobile apps and online services where you can show the screen and a use case;
- service businesses that regularly buy traffic on social media and video platforms;
- teams that need to test offers and segments quickly before an expensive shoot.
Probably not a fit
- premium brands whose value lies in texture, material and genuine emotion;
- products that need to be shown in action in close-up, if there is no existing real footage;
- niches with strict advertising rules (healthcare, finance), where every video goes through lengthy legal review and volume loses its point;
- businesses with a very small ad budget: there is simply nothing to test dozens of creatives on.
Risks: quality, trust and ad disclosure
Quality
AI video can still produce artifacts: unnatural hands, a "swimming" background, lip-sync drift, odd pauses in speech. Every video needs to be reviewed by a person before launch. Viewers forgive a simple picture, but not the feeling of a fake.
Viewers forgive a simple picture, but not the feeling of a fake.
Rights to likeness and voice
If an avatar or voice is based on a real person, you need their written consent for that use, with a clear term and scope. In many jurisdictions a person's image and voice are legally protected, and using them without consent is not allowed.
Ad disclosure rules
Online advertising must be clearly identifiable as advertising. In the EU, for example, platforms have to label ads and show who is behind them, and consumer protection rules prohibit presenting advertising as independent content. An AI video is no different from a filmed one in this respect: if it's an ad, it has to be disclosed as one. Ad platforms handle part of the labeling, but responsibility for accurate information still lies with the advertiser.
Also check the rules of the platforms themselves: some of them require disclosure of realistic synthetic content and run their own ad review.
How it works at Sphere's Content Factory
We packaged the process described above into a separate product, the Content Factory: a subscription to AI videos for performance ads, with no influencers and no shoots. In short, here's how it works:
- Volume and format: 30+ videos a month, 15–30 seconds, vertical 9:16, for Reels, TikTok, Shorts and Stories.
- Launch takes 7 days: we study the product and audience, create an AI brand avatar and a cloned voice, and build the content strategy.
- The first batch of videos arrives 3–5 business days after the content plan is approved.
- What the subscription includes: a content plan, scripts and storyboards, a virality forecast, A/B variants, a monthly call and a chat with a producer.
- The minimum term is 3 months: that's the time needed to go through several cycles of hypotheses, tests and conclusions.
- Free test: a 5-second video in 24 hours so you can judge the quality before you start.
Ad creatives aren't the only place where AI pays off in a company. If you're just starting to use it, it's worth looking at other processes too.
FAQ
Do AI video ads need to be disclosed as ads?
Yes, if the video is used as online advertising. The requirements are the same as for filmed video: it must be clearly marked as an ad and the advertiser must be identifiable. The fact that the video was made by AI doesn't exempt it, and some platforms additionally require disclosure of realistic synthetic content.
Can you make an AI avatar based on one of our employees or our founder?
Yes, if the person has given written consent to the use of their likeness and voice, with clear terms and conditions. Using someone's face or voice without consent is not allowed. The alternative is a fully generated character who becomes the face of the brand.
How many creatives do you need for a test?
It depends on the budget and the number of hypotheses. It helps to think in variables rather than videos: if you're testing three hooks and two offers, you need at least six variants, and each needs enough impressions for a fair comparison. That's why regular performance advertising usually calls for dozens of videos a month.
Will AI videos replace traditional video production?
Not completely. For brand films, premium products and complex scenes, a shoot remains stronger. AI videos win where speed, volume and a constant flow of new variants for testing matter.
How quickly can AI video production get started?
At the Content Factory, launch takes 7 days: product, audience, AI avatar, voice and strategy. The first batch of videos is ready 3–5 business days after the content plan is approved. Before you start, you can order a free test: a 5-second video in 24 hours.




