Summarize with AI
For a hero, video beats a static image only when the motion is the product doing its job: one real input turning into one real output in a short, silent loop. A brand montage of people, cities and logos loses to a plain screenshot, because it costs load time and tells a stranger nothing about what you sell.
This matters most to a founder whose product is hard to explain. You know what your tool does, so a moody montage reads to you as confidence. A visitor who has never heard of you sees a dark rectangle while the file loads, then footage that could belong to any company in any category. The choice is not between moving and still. It is between showing the work and showing a mood.
Three heroes for the same freight tool
Picture an AI tool for freight brokers, used here as an example. A carrier emails a rate confirmation as a scanned PDF. The tool reads it and fills in the load record in the broker's system: pickup, delivery, rate, equipment, and any terms that are missing. Today a coordinator types those fields by hand, about twenty of them per load.
Version one is the hero the company has now. A 9 MB video autoplays behind the headline Freight, reimagined. It runs for thirty-eight seconds: a night time lapse of a highway interchange, a dispatcher laughing into a headset, a map with glowing route lines, a truck at a loading dock at dawn, then the logo. Two buttons sit on top of it, Book a demo and Learn more, both white outline.
Version two is a loop of about twelve seconds. On the left, a crooked scan of a rate confirmation. On the right, an empty load form. The fields fill in one by one as the matching line on the scan highlights: Pickup Joliet, IL, Tue 06:00. Delivery Columbus, OH. Rate $2,450 flat. Equipment 53 ft dry van. Then one field turns amber: Detention terms missing, Ask carrier. A green tag appears, Load 4471 saved, and the loop starts again.
Version three is a single still: the finished load record beside the scan, with three short labels pointing at it. Read from the PDF. Missing term flagged. Saved to your TMS.
Two of these explain the product. One of them only explains the budget.
What a phone shows before the montage plays
Open version one on a phone over cellular data and the first thing on screen is the poster frame, a dark and blurred shot of the highway, with the headline on top. The video is still downloading. In this example, a lab test on a mid-range phone puts the largest paint at a little over five seconds, and for all that time the page says Freight, reimagined over a grey smear.
Google's own guidance is that Largest Contentful Paint should be 2.5 seconds or less, measured at the 75th percentile of page loads, and for a video element it counts the poster image load or the first frame, whichever comes first, according to web.dev's guide to LCP. So the hero file is usually the element being timed. A heavy montage is not a design detail sitting quietly in the background. It is the number.
Then there is what happens when it does play. An iPhone only starts a video on its own when it is muted or has no audio track, and only plays it inline when the video is marked to, as WebKit's video policy sets out. A montage that leans on its soundtrack plays silent, so the visitor watches a truck at dawn with no music and no words. Nothing in thirty-eight seconds of footage names a rate confirmation, a load, or a broker.
Video converts better is the wrong question
Most advice on hero media treats this as a format contest and declares a winner. Video wins because it holds attention, or images win because they load fast. Both camps are answering a question that does not matter. The format is not what converts. What converts is a stranger understanding, in a few seconds, what the thing is and whether it is for them.
Here is the claim worth arguing about: a brand montage is worse than no media at all. A blank hero with a clear headline at least loads fast and lets the words work. A montage adds weight, pulls the eye away from the headline, and fills the space where proof should be with footage that proves nothing. Calling it brand does not change what the visitor sees, which is stock-looking footage of a category, not your product.
The loop and the still both show the work, and that is why they beat the montage. Between those two, the right answer depends on one thing, which the next section turns into a rule.
The rule: does the motion carry the meaning?
Ask what a visitor would miss if you froze the loop on its last frame. If the answer is nothing, use the still. If the answer is the change itself, the moment a messy input becomes a clean output, the loop has earned its weight.
In the freight example, the change is the whole pitch. A broker has seen finished load records all his career. What he has never seen is one filling itself in from a crooked scan, with the missing detention term caught before the truck is booked. That transformation is hard to show in a single frame, so version two earns its place. A product whose value is the finished thing, a report, a ranked list, a clean dashboard, usually does better with version three, which is the same idea as a cropped, labelled product screenshot placed in the hero.
A good example of the motion carrying the meaning is a model demo, where the input and the output are both on the page. The site we built for Chariot, a speech AI lab puts its text to speech model to work on the page: a line you write, such as Flight AI-302 departs T2 at 06:45, next to how the model speaks it, six forty-five in the morning. The point lands because the visitor sees the input and hears the result. No montage could make that case.
If you are weighing the whole hero rather than just its media, the headline and buttons matter as much, and our guide to SaaS hero section design covers them. This post is only the media slot.
The founder who only watches it on office Wi-Fi
The founder of the freight tool approved the montage after two rounds of edits. He has watched it many times, always on a laptop on the office connection, where it starts instantly and the music plays. He has never opened his own homepage on a phone with Wi-Fi off.
On sales calls he never mentions the video. He shares his screen, drags a real rate confirmation into the product and lets the prospect watch the fields fill. That is the moment people lean in. Then the first question on half of his calls is a version of so what exactly does it do, from people who already visited the site. He has rewritten the headline twice this quarter. He has not touched the video, because it looks expensive and it was.
The cost is not the video budget. Every visitor was paid for, in ads, in a post that did well, in his own hours on LinkedIn, and the page fails all of them in the same way. For a sales-led product it also lands on the calendar: demos booked by people who never understood the product, run by the most expensive person in the company. With the median gap from seed to Series A at around 616 days across all sectors, those hours come straight out of runway.
Build the loop backwards from its last frame
If the rule says you need motion, build it in this order. You can do all of it this week without hiring anyone.
- Open your homepage on your own phone with Wi-Fi off. Note what is on screen when the first thing appears. For a lot of visitors, that frame is your hero.
- Run your homepage through PageSpeed Insights on the mobile setting and check which element it names for LCP. If it is the hero video or its poster, the hero is the number to fix.
- Write the transformation in one line: this input becomes this output. If you cannot point to both on a real screen, you do not have a loop yet. Use a still.
- Design the last frame first: the finished output, readable at the size it will show. Export it as WebP or AVIF and use it as the poster. If that frame does not work as a hero on its own, the loop will not either.
- Record the real product, not a mock. Cut every wait. Keep it to one task, around ten to fifteen seconds, with no audio track and no captions needed, so silent autoplay loses nothing.
- Mark the video muted and playsinline, add a visible pause button, and show only the poster to visitors who have reduced motion turned on in their settings.
- On phones, consider showing the poster with a Watch it work button instead of autoplaying. The poster already tells the story. The loop is for the people who want to see it happen.
If you also want a longer walkthrough on the page, keep it out of the hero. A full product video belongs lower down or behind a button, and the demo video guide covers how to order one so the result shows up early.
When the annotated screenshot wins anyway
Even with a good loop, there are cases where the still is the better hero. When the output is a page of text, motion just makes it harder to read. When your product changes every two weeks, a loop goes stale faster than a screenshot and costs more to redo. When the page is mostly visited on phones, a still at the right size is lighter and sharper than any video.
A useful test is five strangers and five seconds. Show each one version two or version three, then hide it and ask what the product does. If both get the same answer, ship the still. If only the loop gets it right, the motion is carrying meaning, and it has earned its bytes. The montage will not pass this test, and that is the clearest way to show your team why it has to go.
A montage is usually approved for a good reason: it makes a small company look like a bigger one. The instinct about stature is right and the medium is wrong, because a stranger reads stature from how quickly the page makes sense. Studio Maydit designs and builds websites for AI founders in the US, UK and Europe, in Framer, Webflow or custom code, and the work runs as a fixed scope of three to four weeks or a monthly retainer with no long lock-in. If you are comparing studios for this, the ranking of landing page design agencies for AI startups lays them side by side. If you want to talk through which of the three heroes your product has earned, book a call with us.
