Meta AI Ad Creation Moves to Generative Video and Agent Tools

From audience buying to asset generation

For most of the last decade, buying ads on Facebook and Instagram was mostly a targeting job. You picked an audience, uploaded creative made somewhere else, and adjusted bids. That model is going away. Meta's announcements through 2026 show a company that wants to make the ads as well as place them. Its own AI systems now produce the images, videos and copy, and they are starting to run the campaign that delivers them.

The clearest recent signal came at Advertising Week New York on October 6, 2026. Meta said that turning static assets into video in Advantage+ creative is now generally available. Its Ads Creative Studio is rolling out to more advertisers after summer tests, and catalog advertisers in a beta can turn product images into product videos at scale23. Meta is also testing agentic features in its business assistant inside Ads Manager. These would let marketers plan a campaign, generate imagery and text, change targeting or move budget by typing a request6.

One way to read this is that ad creation is becoming a feature of the ad platform rather than a separate job that happens before it.

Video is the main target

If there is one theme across the coverage, it is video. Meta says more than 60% of time on Facebook and Instagram is now spent with video, and it uses that figure to explain why it keeps expanding its video generation tools310. The new Ads Creative Studio features go well beyond animating a still image. Advertisers can generate video concepts and variations, adjust hooks, audio and clip length, and add text overlays, captions and end scenes6.

The tools have grown quickly. In mid-2025, around Cannes Lions, Meta was still testing image-to-video ads that added music, text overlays and transitions to static images. It was also trying "video highlights," which pull key scenes out of longer footage25. Industry guides now describe a tool that turns up to 20 product photos into multi-scene video ads, with versions automatically formatted for Feed, Stories and Reels2226. Some guides call the current version "Video Generation 2.0"26, but that label comes from third-party material and Meta hasn't confirmed it.

The money explains the urgency. Meta said in January 2026 that its video generation tools had reached a combined revenue run-rate of $10 billion in the fourth quarter of 20251. One analysis said that figure was growing nearly three times faster than Meta's overall ad revenue23. That makes generative video a significant business line, not a side experiment.

Images get an in-house model

Image generation is changing too. In July 2026, Meta said Muse Image, the first image model from Meta Superintelligence Labs, would power image generation in Advantage+ creative. It promised better photorealism and "product integrity", meaning the advertised product isn't distorted71. That second promise matters. A common complaint about AI ad tools across the industry is warped or low-quality visuals that still need fixing by hand9. For a retailer, a generated image that subtly changes the product is worse than useless.

Meta is also targeting the other common objection to AI creative: that it looks generic. Its June 2026 Cannes announcements included a "brand memory" creative tool (in testing), generation of text inside images, and more translation coverage, with 11 more languages for video voiceover, including Hindi and Arabic7. Coverage of the October update describes Brand Memory as training Meta's AI on an advertiser's colors, logo and approved assets so that variations stay on brand4. Earlier tools had already applied logos, fonts and brand colors automatically at scale2528.

How many advertisers use the tools

This is where the reporting disagrees. Several outlets say more than 4 million advertisers use Meta's generative AI tools, up from 1 million six months earlier822. A July 2026 summary of Meta's Muse Image announcement puts the figure at more than 8 million advertisers using at least one tool7. The most likely explanation is that the 4 million figure dates from late 2025 or early 2026 and kept being repeated, while the 8 million figure is the more recent number. Either way, adoption roughly doubled more than once in about a year. These are no longer early-adopter tools.

Performance claims deserve more caution. Meta says advertisers who follow its assistant's recommendations typically get 5% more conversions at the same cost per conversion3. One analysis noted that Meta published no method for that figure and no list of countries where it applies2. Meta's upgraded ranking and user-understanding models were credited on its Q2 2026 earnings call with an 8.3% increase in ad clicks and a 15.7% lift in conversions on Facebook7. Some vendor blogs go further, claiming a 67% cut in production costs and 18% higher engagement than human-made ads5. Those numbers come from companies that sell automation software, and they should be treated as marketing.

AI-generated "UGC"

The least discussed part of this shift is AI-generated user-generated content. UGC-style ads work on Meta because they look like something a real person posted, not a brand spot, and they fit into Reels and Stories without setting off the viewer's urge to skip14. The problem has always been volume. Real creators are slow to brief, film and revise, and that makes heavy testing expensive1418.

Third-party tools now sell synthetic presenters to fill that gap. One service offers 16 AI "creators" and delivers a vertical video with a voiceover in about five minutes for a flat $611. Others advertise libraries of more than 1,000 AI actors built on many video models16, or dozens of hook and avatar variations generated in a single session19. Meta's own tools mostly don't make synthetic people. They work from the advertiser's existing images and product assets. But the platform's appetite for constant creative variation is what creates the market for these outside tools. Meta's ads MCP server makes the connection tighter. It now lets outside AI tools upload creative, preview ads and send agent-built work into Ads Manager as drafts. Ten partners have integrated it so far62.

Practitioners describe a split approach. A common pattern in industry advice is to use AI UGC for cheap, high-volume testing, then pay real creators to polish the angles that work14. One advertiser on a Reddit forum said AI UGC works for reaching new audiences but tends to fail in retargeting, because people who have already seen a brand are "harder to fool"15. That comment says a lot. Synthetic UGC works by looking like a real recommendation, so its value depends partly on viewers not noticing it isn't one.

That creates a policy risk. One compliance guide says Meta judges AI ads under the same Advertising Standards as other ads. It lists rejection triggers such as presenting an AI presenter as a real customer testimonial, or using a synthetic likeness of a real person20. Another vendor says ordinary e-commerce ads with AI avatars don't currently need special disclosure outside political contexts14. The two accounts don't fully agree, so the rules are unsettled. The compliance guide itself warns that Meta has revised its approach several times20. Meta's consolidated ad terms take effect on October 30, 2026, and they fold the Ad Creative Generative AI Terms into the Self-Serve Ad Terms. Advertisers using generative tools should read that document47.

The goal: a URL and a budget

The direction of all this is not in doubt. Reporting by The Wall Street Journal described Meta's plan to let a brand supply a product image and a budget goal, with AI producing the visuals, video and copy and choosing who to target9. Zuckerberg called it "a redefinition of the category of advertising"9. Newer accounts describe the target input as a product URL plus a budget, with AI building the whole campaign across Facebook, Instagram, Messenger and WhatsApp822. Comparisons with competitors put Meta at the most aggressive end. Google's Performance Max is described as assembling some assets, while TikTok's Smart+ is described as relying on human-made assets27. That comparison is debatable, since another analysis credits TikTok's Symphony with generating complete video ads from a text prompt23.

The October announcements suggest Meta is close to that goal but hasn't reached it. The creative-generation pieces are live or in beta. The pieces that run campaigns on their own are still tests that require advertiser approval before anything goes live62.

What it means

The likely result is a reshuffling of creative work rather than its end. Producing ads is becoming cheap and plentiful. Creative direction, brand guidelines and independent measurement are becoming the scarce skills. Meta now shows a creative diversity score in Ads Manager, which suggests its systems reward genuinely different concepts over near-copies7. Measurement is the unresolved problem. One analysis cites 640 incrementality tests by the measurement firm Haus. Those tests reportedly found Advantage+ campaigns underperforming manual ones over time, despite strong results in Meta's own reports24.

The main risk for advertisers is not that AI-made ads will look bad. It is that when one company makes the ad, places it and reports how well it worked, the advertiser loses the ability to check the results independently. The tools are improving quickly, but advertisers who keep their own measurement will be in the best position to judge whether they are working.