Short answer: a Facebook video ad that converts is vertical (1080x1920, 9:16) for Reels and Stories or 1080x1350 (4:5) for the feed, runs 10 to 15 seconds, is built for sound-off with burned-in captions, and lands its hook in the first second. The formats that win are UGC talking-head, hook-plus-demo, product-in-use, and problem-solution, because they feel native and show the product working. But the real lever is not any single spec or format. It is how many videos you can produce and test, because winning creative is found, not designed.
The rest of this page is the why, the failure modes, and the decisions that move CPM and cost per result in 2026.
Roughly 600 people search for "facebook video ads" in the US every month, and most of them want one of two things: the exact specs so their upload stops getting cropped, or the creative formula that makes the thing actually sell. This guide covers both, and then the part nobody puts on a spec sheet, which is the production-cost ceiling that quietly caps how good your account can ever get.
Why video is the default format on Facebook now
Static still has a job, and it is a real one. But video is where the cheap, scalable inventory lives in 2026. Reels alone reshaped the auction: vertical, full-screen, sound-on-by-default, swipeable, and effectively infinite. Meta's delivery system, now powered by the Andromeda retrieval model, can only show your video where you have given it native creative to show. If you upload one square clip, you are competing for a sliver of the inventory. If you upload vertical, square, and feed-native cuts, you let the algorithm find the cheapest impressions across every placement.
That is the strategic frame for everything below. The specs are not cosmetic compliance. They decide which inventory you are even eligible for, and which inventory is the cheapest. Get the format wrong and your CPM is structurally higher before a single person has decided whether they like your ad.
The Facebook video ad specs that matter
Here is the cheat sheet, worth pinning above your edit bay. These are the 2026 recommended specs by placement, not the absolute minimums Meta will technically accept.
| Placement | Aspect ratio | Recommended size | Max length | Notes |
|---|---|---|---|---|
| Feed | 4:5 (also 1:1) | 1080 x 1350 | 241 minutes | 4:5 takes the most vertical real estate in-feed; 1:1 is the safe fallback. Most people stop here, so the first second decides everything. |
| Reels | 9:16 | 1080 x 1920 | Up to 90 sec for direct response | Full-screen, sound-on-by-default, the cheapest scalable inventory. Keep text and CTAs out of the top/bottom UI safe zones. |
| Stories | 9:16 | 1080 x 1920 | Up to 60 sec | Tap-through and skippable. Front-load everything; assume a 1 to 3 second view before a thumb decides. |
| In-stream | 16:9 | 1920 x 1080 | 5 to 120 sec | Mid-roll inside longer video. Skippable after a few seconds, so the hook still has to do the work fast. |
Across all of them: MP4 or MOV container, H.264 video codec, AAC audio, under 4GB, 30fps and up. If you build only one asset, make it 1080x1080 square, which renders acceptably nearly everywhere, but understand that you are leaving screen space and cheap vertical inventory on the table.
The variant question buyers ask most often, "what facebook video ad size do I actually need," has a clean answer for 2026: build vertical-first at 1080x1920, then resize to 4:5 and 1:1. Resizing down from vertical preserves your composition. Stretching up from square does not.
Facebook video ad length: how long should it be
The data and the practice both point to the same place for direct response: 10 to 15 seconds. That window is long enough to hook, show the product doing its job, and ask for the click, and short enough that a meaningful share of people watch to the end. Completion rate matters because Meta reads it as a quality signal, and because the people who finish are the ones who saw your offer.
That is not a universal rule. Three caveats:
- Talking-head and problem-solution ads routinely run 30 to 60 seconds and still convert, because the format is built on a narrative that needs room. The rule that survives is not "stay under 15 seconds," it is "earn every second after the first."
- Reels and Stories reward slightly longer in 2026 than they used to. A 20 to 30 second Reel that holds attention can outperform a 10 second cut that says nothing.
- In-stream is its own animal. You have a captive but impatient viewer, so the first 5 seconds before the skip option are the whole game.
The practical move: launch your test at 10 to 15 seconds, find the winner, then test a longer variant of that exact winner. Do not start long. Starting long is how you burn budget proving that nobody watched past second three.
Build for sound-off, then add sound as a bonus
A large share of feed video is watched on mute. Reels lean sound-on, but the feed does not, and you do not get to choose where a given impression lands. So the safe assumption is that your message has to survive with the sound completely off.
That means burned-in captions, not auto-captions you hope render. It means the story is legible from the visuals alone. It means any spoken hook also appears as on-screen text in the first second. Then, when sound is on, the audio is a bonus that lifts the experience rather than a dependency that breaks the ad when it is missing.
Mute your own ad and watch it. If you cannot tell what it is selling, neither can the feed.
The first-second hook rule
This is the single highest-leverage thing on the page, so it gets its own section. The thumbstop happens in the first second. Not the first three, not the first five. One.
The metric to watch is the thumbstop ratio, sometimes called hook rate: the share of people who watch past the first three seconds. If that number is weak, nothing downstream matters, because almost nobody saw the downstream. A strong hook does one of a few things in second one:
- Names the problem the viewer already has ("If your Facebook video ads stopped converting...").
- Shows an unexpected or pattern-breaking visual.
- Makes a specific, slightly provocative claim.
- Opens mid-action, with the product already doing something.
What kills the hook: a logo intro, a slow brand sting, a wide establishing shot, a person taking a breath before they start talking. Cut all of it. The hook is the ad. For a deeper read on which creative signals actually drive performance, creative analytics is the discipline that connects hook rate to spend.
The Facebook video ad formats that convert
Format is the structure of the creative, not the spec. These four out-perform polished brand films for direct response, consistently, across accounts. Use this as your testing menu.
| Format | What it is | When to use |
|---|---|---|
| UGC / talking head | A real-feeling person talking to camera, phone-shot aesthetic, often holding or using the product. | Default for direct response and app installs. Feels native, builds trust fast, scales into many variants cheaply. |
| Hook + demo | A sharp first-second hook followed immediately by the product doing the thing it does. | When the product's value is visual and obvious in a few seconds. Great for software, tools, and physical products with a clear "aha." |
| Product in use | The product shown in its real context, doing its job, minimal talking. | When seeing it used answers the buyer's main objection. Strong for ecommerce, home, and lifestyle. |
| Problem - solution | Open on the pain, agitate briefly, present the product as the fix, close with the click. | When the audience is problem-aware but not yet sold on your specific answer. The workhorse of paid social. |
The honest part: you will not know which of these wins for your product until you run them. The "best facebook video ads" in any account are not the ones a creative director loved. They are the ones that survived the auction. That is why the list above is a testing menu, not a ranking. You run all four, find the one or two that win for your audience, then make ten variations of the winner.
The mistakes that quietly tank video ad performance
These are not catastrophic failures that throw an error. They are the silent leaks that make a perfectly fine ad underperform a great one, and most accounts have at least two of them running right now.
| Mistake | What actually breaks | Fix |
|---|---|---|
| Uploading one square asset everywhere | You are eligible for a sliver of inventory, so delivery defaults to the most expensive placements and your blended CPM rises. | Build vertical 9:16, then resize to 4:5 and 1:1. Give the algorithm native creative for every placement. |
| Sound-dependent message | On mute, the ad sells nothing, so a large share of impressions are wasted. | Burn in captions, make the visual self-explanatory, mute-test before launch. |
| Slow intro | Logo stings and establishing shots burn the one second that decides the thumbstop, so hook rate craters. | Cut to the hook. The first frame should already be the most interesting thing in the ad. |
| Testing one or two videos a month | You cannot find a winner in a sample of two, and you cannot outrun creative fatigue. | Raise production volume to 10 to 40 assets a month so the auction has enough to choose from. |
| Editing winners to death | Every edit to the ad set can re-trigger the learning phase and reset performance. | Duplicate to iterate. Leave the proven ad set stable and test new creative beside it. |
| Treating fatigue as an audience problem | You widen the audience when frequency climbs, but the audience was never the issue, the creative was tired. | Refresh creative on a cadence. New videos beat new audiences for a tired ad. |
The thread running through all six: the cost of being wrong on video is rarely a hard failure. It is a structurally higher cost per result that you never notice because the ad still "works," just worse than it should.
Where each format runs and how much it costs you
A quick orientation on placement economics, because it changes how you should weight your production.
Reels is, in 2026, the cheapest scalable video inventory on the platform. It is vertical, full-screen, sound-on, and effectively unlimited, which keeps CPMs lower than the feed for comparable audiences. If you are optimizing for cost per result and your creative is vertical-native, Reels should be carrying a large share of your video spend.
The feed is more expensive per impression but higher-intent in many verticals, and 4:5 video there takes more screen than 1:1. Stories sits between the two, cheap and fast, but skippable in a heartbeat, so it rewards front-loaded creative more than any other placement. In-stream is a niche, useful for reach and for audiences who tolerate mid-roll, but it is not where most direct-response budgets should concentrate.
The practical implication: do not let placement be an afterthought you check at the ad-set level. Let it shape what you produce. If Reels is where your cheap inventory is, vertical 9:16 should be your default build, not your resize. For the full placement-by-placement spec breakdown across Facebook and Instagram, see the Meta ad sizes reference.
The real problem: producing enough video to test
Everything above is solvable with a good editor and a checklist. This is the part that is not.
Winning creative is found, not designed. On cold traffic, the share of new videos that become winners is low, often in the single digits to low teens. That math has one inescapable consequence: to find winners reliably, you need volume. If you produce one or two videos a month, you are running a lottery with two tickets. You will occasionally win, and you will spend most of your time confused about why an account that "does everything right" plateaus.
Traditional video production caps that volume hard. A UGC shoot means briefing creators, waiting on delivery, paying per asset, and editing. The cost per finished video runs from tens to hundreds of dollars and the calendar runs in weeks. That production-cost ceiling is the real constraint on most accounts. It is not targeting, it is not bidding, it is not the pixel. It is that you cannot make enough video to test your way to a winner.
This is also why creative fatigue is so punishing. The same audience sees the same small pool of videos, frequency climbs, CTR drops, and your effective CPM rises. The fix for fatigue is always fresh creative on a cadence, and if your cadence is one video a month, you cannot keep up. The production ceiling and the fatigue problem are the same problem wearing two hats.
How autonomous AI marketing agents handle video production at volume
The reason the production ceiling has held for so long is that making video was the expensive, slow, human-bottlenecked step. Autonomous AI marketing agents remove that step, which is what changes the math.
Superscale is an agentic AI Ad Agent, not a clip generator. You paste a link, an App Store page, a Shopify store, a website, and the Agent researches your product, your competitors, and the top ads running in your niche, then produces 10 or more launch-ready video and static ads in minutes. It draws on 300+ AI UGC characters, 25+ languages, and best-in-class lip-sync, so the UGC talking-head and problem-solution formats that convert are exactly the ones it can spin up at volume. Agent video runs 15 seconds, Sora 2 up to 20 seconds, and Speaking Character templates up to 120 seconds, which maps cleanly onto the length guidance above. Because it connects directly to your Meta, TikTok, and Google ad accounts, it can read account data, build new variants, iterate on winners, and pause underperformers, which is the fatigue-refresh loop running on its own. The built-in Competitor Tool surfaces what your competitors are running via the Meta Ad Library so your testing menu starts from what already works in your category.
The number that makes the case concretely: StromNow went from one video a week to ten, doubled app installs, and cut their cost per video roughly 20x, from over $100 to about $5, producing 40+ AI-UGC assets a month at around 15 minutes per asset. That is the production ceiling lifting. When a fresh, on-format video costs $5 instead of $100, testing thirty of them a month stops being a budget decision and starts being a default.
If you want the broader argument for whether this can replace a human media-buying motion, can AI replace a media buying team walks through where agents take over and where they do not. The honest version: agents do not replace taste, they remove the production tax that kept good taste from being tested at scale.
A simple decision framework: which video to make next
When you are staring at an account and do not know what to produce, run this in order.
- Do you have a vertical 9:16 asset for every active concept? If not, that is the first build. You are almost certainly under-serving Reels and Stories inventory.
- Is anything winning right now? If yes, make ten variants of the winner before you make anything new. Iterating on a proven hook beats inventing a new one.
- Is everything fatiguing at once? That is a volume problem, not a concept problem. You need more fresh assets per week, not a clever new audience.
- Are you testing fewer than four distinct formats? Run all four from the table above. You cannot know your best format until each has had real spend.
- Is production the bottleneck on every answer above? If "I can't make enough video" is the reason you are stuck, that is the constraint to solve, with ad creative automation or an agent, before you touch targeting or bidding again.
The framework collapses to one idea: spend your next hour on whatever is currently capping your test volume, because test volume is what finds winners.
FAQ
What is the best length for a Facebook video ad?
For most direct-response campaigns, 10 to 15 seconds. Long enough to hook, show the product working, and ask for the click; short enough that people watch to the end. Talking-head and problem-solution ads can run 30 to 60 seconds and still convert, but only if the first second earns the rest. Start short, then test longer versions of your winners.
What are the specs for a Facebook video ad?
Use 1080x1920 (9:16) for Reels and Stories, 1080x1350 (4:5) for the feed, and 1080x1080 (1:1) as a safe cross-placement fallback. Keep video under 15 seconds for direct response, design for sound-off with burned-in captions, and put your hook in the first second. MP4 or MOV, H.264, under 4GB.
What size should a Facebook video ad be?
The native vertical size is 1080x1920 pixels at a 9:16 aspect ratio for Reels and Stories. For the feed, 1080x1350 at 4:5 takes the most screen real estate. If you build only one size, 1080x1080 (1:1) renders acceptably almost everywhere, but it leaves screen space on the table in vertical placements.
Do Facebook video ads need captions?
Yes. The majority of feed video is watched with sound off, so any spoken message has to survive on screen as text. Burn captions into the video rather than relying on auto-captions, keep on-screen text inside the safe zones, and make sure the ad still makes sense and still sells with the sound completely muted.
What kind of Facebook video ads convert best?
UGC-style talking-head ads, hook-plus-demo, product-in-use, and problem-solution structures consistently outperform polished brand films for direct response. They feel native to the feed, they show the product working, and they open with a clear hook. The format matters less than the volume of formats you can test.
How many video ads should I test?
More than you think, and more than you can usually afford to shoot. Winning rates on cold creative are low, so volume is the lever. Teams using AI to produce video routinely run 10 to 40 fresh assets a month instead of one or two, which is why their cost per winning ad drops. One Superscale customer cut cost per video roughly 20x, from over $100 to about $5.
Why do my Facebook video ads stop working?
Creative fatigue. The same audience sees the same video too many times, frequency climbs, CTR drops, and your effective CPM rises. The fix is not a new audience, it is new creative on a steady cadence. If you can only make one video a month, you cannot outrun fatigue, which is why production volume is the real constraint.
Are vertical or square video ads better on Facebook?
Vertical (9:16) wins where most cheap, scalable inventory now lives, which is Reels and Stories. Square (1:1) is the safe generalist that renders fine everywhere. The 2026 best practice is to build vertical-first and resize down, rather than build square and stretch up, because Meta's delivery favors placements you have native creative for.
