The fastest way to expose an AI-made landing page is to make it obvious that the AI made every aesthetic decision for you. Same layout as always, same artificial lighting, same static hero, same polished-but-pointless finish. The result can look clean, sure, but it also looks interchangeable.
The more interesting approach, at least to me, is something else entirely: don’t use AI to spit out a whole page. Use it to generate one specific visual asset that you then control properly on the frontend. In this case, a product video that starts with the object exploded into parts, assembles on camera, and then becomes the core of a landing page with scroll reveal. At that point AI stops being “the full designer” and becomes one piece inside a much clearer creative direction.
That’s exactly why this format works so well. Instead of asking AI for a finished website, you ask for a very specific clip: 16:9, 9 seconds, 1080p, with a product floating in a studio and a sequence where the parts line up and end up forming the final object. Then that video doesn’t play like a normal ad. You use it as a visual timeline and control the frames with scroll. The feeling changes completely. It no longer looks like the usual generic landing page generated in one shot. It feels like an experience that was actually designed on purpose.
The underlying reason is simple: most AI-made landing pages fail because they try to solve visual identity, copy, structure, motion, and hierarchy all at once. Too many delegated decisions, not enough taste. The trick here is to narrow the problem down. AI generates the base product motion and you handle the pacing, the layout, when the text appears, and how the navigation feels.
If you want to see a real coded example of this idea, the source repo is here: edunavajas/custom-landings. The pattern becomes very clear there. There are examples with a phone, keyboard, and camera, and the interesting part is not just the visual finish but the logic: a full-screen hero video, a sticky section, and scroll scrubbing that reveals the product progressively.
Another important detail: you usually should not use the AI-generated clip exactly as it comes out. For scroll scrubbing, seeking needs to feel clean. If the file is compressed badly or has a keyframe structure that is not scrub-friendly, the transition feels dirty, jumps around, and kills a lot of the effect. That is why it makes sense to convert it first with ffmpeg and prepare a version that behaves better for scrubbing. A reasonable command would be this:
ffmpeg -i input.mp4 -c:v libx264 -preset slow -crf 18 -g 1 -pix_fmt yuv420p -movflags +faststart output-seekable.mp4The point is not to “optimize for the sake of optimizing”, but to prepare the clip so scroll can move through frames more smoothly. In the repo you can already see that pattern, with a separate file prepared specifically for this use case.
Once the technical side is clear, what really matters is the prompt. If you want the result not to smell like AI, the prompt has to describe the product, composition, materials, lighting, and motion with intent. “Make me a modern landing hero” is not enough. You need to ask for a shot that works as a cinematic base, not a generic demo.
If you want to generate that kind of product clip faster, I’d also test it with tools built for this kind of workflow. This Higgsfield referral link fits well here because the whole point is the same: get a good base asset first, then build the landing around it instead of expecting the AI to solve the entire website for you.
Here are the prompts from the example in English, exactly the way it makes sense to use them as a starting point for the visual asset:
Prompt for smartphone
A premium futuristic smartphone floating in a dark luxury studio, exploded view with all internal parts separated in a clean and elegant composition, then smoothly assembling into the final product. Cinematic lighting, ultra detailed materials, polished metal edges, premium glass reflections, centered composition, black background, high contrast, realistic shadows, modern product commercial aesthetic, 16:9, 9 seconds, 1080p.Prompt for keyboard
A premium mechanical keyboard floating in a dark cinematic studio, exploded into separate keys, switches, plate and body, then smoothly assembling into the final keyboard. Elegant product commercial lighting, ultra detailed materials, brushed metal, subtle reflections, centered composition, black background, realistic shadows, modern luxury tech aesthetic, 16:9, 9 seconds, 1080p.Prompt for camera
A premium professional camera floating in a dark luxury studio, exploded view with lens, rings and body parts separated in a clean cinematic composition, then smoothly assembling into the final camera. Dramatic soft lighting, ultra detailed metal and glass materials, realistic reflections, centered composition, black background, high contrast, modern product commercial aesthetic, 16:9, 9 seconds, 1080p.The nice part is that all three follow the same logic: suspended object, separated parts, clean assembly, premium ad aesthetic. Then when you take that into the frontend, the hero stops depending on a static render and gains a lot more presence with very little. Scroll, video, measured overlays, short copy. You do not need twenty extra animations around it for it to work.
If you look closely, the real value of this approach is not “using AI”, but using AI for one very specific thing and then building a web experience on top that feels intentional. That is the difference between a landing page that feels like just another template and one that, even with AI-generated assets, feels modern, controlled, and much harder to identify as “made by AI”.
If you want to replicate it, I would not start with the full website design. I would start with the right asset: a good short horizontal video with a clear product assembly. Then build the scroll reveal and only after that write the rest of the landing page around that motion. In this case, the build order matters a LOT.
What do you think?
Leave your opinion, question or suggestion. Comments are synced with GitHub Discussions .