# How to make a product launch video: a step-by-step guide How to make a product launch video, step by step, with real lengths, structures and beat timings from 66 launch videos, plus where AI tools fit. Published 2026-11-11. https://motionreference.com/blog/how-to-make-a-product-launch-video A product launch video has about 45 seconds to explain something new, make it look worth trying, and tell people where to get it. This guide walks through how to make a product launch video in eight steps, from the one claim it has to land to the frame it ends on. The numbers come from the 66 product launch videos in the Motion Reference library (as of November 2026), each analysed scene by scene: their structures, their lengths, and how their beats are timed. They describe what launch teams actually ship, not rules. Use them as a starting point and break them on purpose. ## 1. Decide the one thing the video has to prove Every strong launch video proves one claim. Write it as a sentence a viewer could repeat: "it transcribes two languages in the same sentence", "your agent can now run while you sleep", "one image becomes a whole campaign". Then decide how you'll prove it on screen. A claim with no visible proof becomes a title card, and title cards are easy to scroll past. If you can't film or capture the proof yet, fix that before you write a script. ## 2. Pick a structure The structure is the shape of the video: what the viewer sees first, and what they're waiting for. In the library's 66 product launches, three shapes cover most of them: | Structure | Videos | Typical length | Use it when | | --- | --- | --- | --- | | Hook → reveal | 23 | 30s | The product changes how something is done, and curiosity carries the opening | | Feature montage | 22 | 53s | You're launching several capabilities and each can be shown working | | Problem → solution | 10 | 65s | The audience has a specific, familiar pain | | Manifesto | 4 | 68s | You're selling a point of view as much as a product | | Transformation | 4 | 23s | Range is the selling point: one input, many outputs | Hook → reveal opens on a question, then answers it with the product. Runway's Solaris spends its first 11 seconds on archival footage of fixed, old software before revealing a generated interface: [Solaris: Interface World Model by Runway](https://motionreference.com/reference/2e01b053-461f-4b8f-80f0-1d0a88b8b1cf) (Markdown: https://motionreference.com/reference/2e01b053-461f-4b8f-80f0-1d0a88b8b1cf.md) Problem → solution needs time to make the problem felt, which is why it runs longest. Docker's Sandboxes opens on the friction of supervising coding agents, then shows the fix: [Introducing Docker Sandboxes by Docker](https://motionreference.com/reference/ff920200-6a8c-461d-871c-3cab3c4cdd58) (Markdown: https://motionreference.com/reference/ff920200-6a8c-461d-871c-3cab3c4cdd58.md) For a closer look at each structure, browse the library by structure and watch three or four of the same shape back to back. ## 3. Set the length The typical product launch video in the library runs 48 seconds. Half run between 29 and 66 seconds. 17 are under 30 seconds, and 8 run 90 seconds or longer. Let the structure set the length, not the other way round. A hook → reveal can land in half a minute. A problem → solution needs a full minute because the problem has to be felt before the fix means anything. Length should come from proof: every extra 15 seconds should prove something new. ## 4. Plan the beats A beat is one section of the video with one job. The typical launch video in the library has 5 beats. The first and last each run about 8 seconds, which leaves the middle for proof. A 45-second outline that fits most launches: | Time | Beat | Job | | --- | --- | --- | | 0:00–0:08 | Hook | Earn the next ten seconds: the problem, a question, or the product doing something surprising | | 0:08–0:14 | Reveal | Name the product and its one-line promise | | 0:14–0:30 | Proof | Show the claim working, in the product, start to finish | | 0:30–0:37 | Range | One or two more uses, faster | | 0:37–0:45 | Close | Logo, availability, where to get it | Close on a clear end card. 54 of the 66 product launches in the library end on a logo or call-to-action beat. Say where it's available (a URL, "live now", the platforms) so nobody has to guess. ## 5. Write the words: on-screen type or voiceover Most launch videos don't use a voiceover. In the library, 43 of the 66 product launches are music-led, 13 use a voiceover and 6 are driven by sound design. Kinetic typography (words animated on screen) appears in 49 of them. On-screen type suits short videos and silent autoplay feeds: one phrase per beat, timed to the cut. A voiceover suits videos that explain a mechanism or follow a founder. ElevenLabs' Scribe v3 uses one to walk through benchmarks and live demos over two minutes: [Introducing ElevenLabs Scribe v3 and Scribe v3 Live by ElevenLabs](https://motionreference.com/reference/e12a99bf-923c-49a4-ad2c-f5e968f2921c) (Markdown: https://motionreference.com/reference/e12a99bf-923c-49a4-ad2c-f5e968f2921c.md) Whichever you choose, write the words before you design anything. Read them aloud against the beat timings. If a line doesn't fit its beat, cut the line, not the beat. ## 6. Choose the visuals Show the real product. Product UI appears in 48 of the 66 launches in the library, typography in 58, live action in 30 and 3D in 28. UI choreography (the interface animated on screen, with a cursor and real interactions) appears in 43. A few patterns that recur in the library: - **Start each feature with the action that triggers it**: the prompt, the click, the command. The viewer sees cause, then effect. - **Keep one example throughout.** Ideogram's 4.5 launch follows one perfume campaign from shoot to billboard instead of showing unrelated samples. - **Show the "before" with real footage** when the product replaces an old way of working. - **Turn the invisible into a character or a diagram** for infrastructure and technical products. Docker gives its agent a mascot; Replit draws branching and merging. ## 7. Where AI tools fit AI video models can now make convincing shots, but a launch video still comes together shot by shot. A single generation is short: Google's [Veo 3.1](https://docs.cloud.google.com/vertex-ai/generative-ai/docs/models/veo/3-1-generate) makes clips of 4, 6 or 8 seconds, [Gemini Omni 1.2 Flash](https://aistudio.google.com/docs/omni) up to 15, [Kling 3.0](https://kling.ai/feature/kling-video-3) up to 15, and Luma's [Ray3.2](https://lumalabs.ai/ray3-2) 5 or 10. **Where AI tools help:** - **Hooks, atmosphere and B-roll**: the abstract or cinematic shots around the product. - **Concept frames before you commit**: generate stills for each beat, approve them, then animate. Runway's [product video guide](https://dev.runway.com/learn/generate-product-videos) recommends exactly this order: "Write the script first, then generate the key frames, then generate video from those frames." - **Versions**: other aspect ratios, audiences and markets. In IAB's 2025 survey, making versions for different audiences was the most common planned use of generative AI in video ads ([IAB](https://www.iab.com/news/nearly-90-of-advertisers-will-use-gen-ai-to-build-video-ads/)). **Where real material still matters:** - **On-screen text and logos.** Luma's own prompt guide says: "[Never ask AI to render readable text or logos](https://lumalabs.ai/news/write-ai-video-prompts)." Runway warns that a small label "comes back as gibberish text" and that text on a product should come from your reference images. An October 2026 benchmark of 11 video models found that even the best got about one word in four wrong ([VTR-Bench](https://arxiv.org/abs/2610.01499)). Add titles, type and logos in the edit. - **Your product's interface.** No vendor offers guidance on generating product UI, and a launch video is a promise about what people will see. Our advice: screen-capture the real thing and animate it in the edit. - **Your product's exact look.** Give the model several angles of the real product. Runway again: "One hero shot is not enough." The beat plan from step 4 is your shot list. For prompts written shot by shot from real launch videos, see [how to prompt AI video tools for a product launch video](/blog/ai-video-prompts-product-launch). ## 8. Sound, pace and the edit Pick the track before you cut, then cut to it. 30 of the 66 product launches in the library are cut at a fast pace and 10 escalate, building speed toward the end. Rapid cutting appears in 37. Silence is a choice too. Sound design (clicks, whooshes, keyboard sounds) can carry a video on its own, and it makes interface moments feel physical. Export landscape first: 64 of the 66 product launches in the library are landscape. Then check each platform's current specs for vertical and square cuts before you reframe. ## A checklist before you ship - [ ] One claim, proved on screen - [ ] A structure chosen on purpose, and a length that fits it - [ ] A hook in the first 8 seconds - [ ] The product named by the second beat - [ ] Every feature opens with the action that triggers it - [ ] Words work with the sound off - [ ] Every shot of the product is real, or clearly labelled as illustrative - [ ] An end card that says where to get it ## FAQ ### How long should a product launch video be? The typical product launch video in the Motion Reference library runs 48 seconds, and half run between 29 and 66 seconds. Hook → reveal videos tend to be shortest (about 30 seconds) and problem → solution videos longest (about a minute). ### What's the best structure for a product launch video? There isn't one best structure, but three cover most launches: hook → reveal (23 of 66 in the library), feature montage (22) and problem → solution (10). Choose by what you need to prove. ### Do I need a voiceover? Usually not: 43 of the 66 product launches in the library are music-led, with on-screen type carrying the message. Use a voiceover when there's a mechanism to explain or a person to follow. ### Can I make a product launch video with AI? Partly. See step 7 for where AI video tools help and where real screen capture still matters. --- Find examples for your own launch in the [Motion Reference library](/): search by structure, technique or industry, or connect your AI agent over [MCP](/connect) to plan a video from real references.