How to Make a Product Launch Video with AI (Three Real Runs, With the Briefs)

A product launch video has one job that no other video has: it has to exist on launch day. Miss that and the best cut in the world is a post-mortem. This guide is the workflow I actually use to get one out the door in an afternoon: describe the launch in plain language, let AI build the motion, fix it in plain language. Every example below was generated for this article — first take, flaws left in — with the exact brief shown so you can copy it.

If you want the theory of what a launch video is, that's covered in the product launch video glossary entry; if you want to see finished ones by situation, the examples article has four. This one is the how.

Step 1 — Decide what kind of launch this is

Launch videos split into two families, and the family decides everything downstream:

• Text-led. A feature drop, an app going live, a beta invite, a rebrand. There is no physical object to film; the video is made of type, motion, and UI. These run on iArt's fast engine — end to end, a finished clip landed in two to three minutes in every run below.
• Photo-led. A physical product: a serum, earbuds, a kettle. You attach a product photo and the film is built around the real object. These take longer (about six minutes in my run) and cost more.

Pick the family first. Mixing them — asking for a physical product reveal with no photo — is the most reliable way I've found to get a video that looks generic: the engine has nothing to hold on to, so it invents a plausible product.

Step 2 — Write the brief as a shot list

In my runs, the difference between a launch video that lands and one that meanders has been the brief. A working brief is not a mood board; it's a shot list with times on it. The shape that works:

• Format and length first: "15-second launch teaser, 16:9" or "20-second app teaser, 9:16 vertical."
• Timecoded beats: 0–3s the hook, 3–7s the reveal, 7–12s the proof (features, screens, details), 12–15s the name and the call to action.
• Exact on-screen words. Give the headline, the feature names, the CTA — and say that those are the only words you want. A closing constraint like "no small print anywhere" helps, but as the first clip below shows, it isn't a guarantee.
• One palette sentence: background, type color, one accent.

Here is the brief for a feature launch, exactly as submitted:

A 15-second launch teaser for a new AI search feature in a note-taking app called Lumen, 16:9. 0-3s: dark screen, a single cursor blink, the words "Your notes" in big kinetic type. 3-7s: "now answer back" lands word by word, then an animated UI mockup slides in showing a question typed into a search bar and an answer assembling below it. 7-12s: three feature beats in bold type, one per beat: "Ask anything", "Cites your notes", "Works offline". 12-15s: Lumen wordmark, tagline "Search that reads what you wrote", CTA "Try it free". Palette: near-black background, warm white type, one electric-violet accent. No small print anywhere.

And what that brief produced, first take:

Lumen (a fictional app) — feature launch teaser, 15 seconds, generated in 2 min 28 s. Every beat is there, in order. It also added a line of small-print sub-caption under each of the three features that the brief had explicitly banned.

Step 3 — Generate, then read the result against the brief

Don't judge the first take on taste yet. Judge it on compliance: did every beat land, in order, with the words you specified, and nothing else? That's the check that tells you whether the brief was good. The Lumen clip lands every beat — hook, reveal, UI mockup, three features, wordmark, CTA — and then breaks the last line of the brief by writing a sub-caption under each feature. So: beats pass, constraint fails, and "remove the sub-captions" is the first sentence of take two. If a beat is missing or extra, the fix is to make that line of the brief more literal, not to add adjectives.

Two more launches from the same session, to show the range one workflow covers. An app going live, vertical for social:

A 20-second app launch teaser for a habit-tracking app called Streak, 9:16 vertical. 0-4s: countdown 3, 2, 1 in huge type, each number lands with a bass hit and a ring pulse. 4-8s: the app name "Streak" and its icon (a single orange flame) reveal from the center. 8-15s: two phone screens slide in one after the other: a daily check-in screen, then a calendar heat-map of completed days. 15-20s: "Build the habit. Keep the streak." then "Available now on the App Store". High energy, punchy cuts, deep navy background, orange accent. No small print.

Streak (fictional) — app launch, 20 seconds, 9:16. Countdown, icon reveal, two screens, App Store close.

And a beta invite, where the whole video is typography:

A 15-second beta invite film for a design-collaboration tool called Canvasly, 16:9. 0-3s: bold headline "You're early." on an off-white background, editorial serif. 3-11s: three benefits in kinetic type, one at a time, each with a small abstract shape animating beside it: "Comment on anything", "Version history that reads like a story", "Handoff without the handoff meeting". 11-15s: "Join the beta" CTA with a subtle underline draw-on, Canvasly wordmark below. Confident, a little dry, minimal motion. No small print.

Canvasly (fictional) — beta invite, 15 seconds. The brief asked for "a little dry"; the result is restrained, as asked.

What the three cost, from the account ledger: $0.86 in total for 15 + 20 + 15 seconds of video — about 29 cents each on average. Wall-clock: 2:28, 2:36, and 2:02.

Step 4 — Physical product? Attach the photo

For a physical launch the brief changes in one way: you attach the product photo and tell the engine what must survive from it. Before rendering, iArt's own agent warns that the label wordmark is redrawn rather than copied pixel-for-pixel, so keep it large and frontal to give it the best chance. The brief I used for a serum launch (note that I broke my own rule and asked for a small line of text):

A 10-second product launch reveal for this NOVA vitamin C serum, 16:9. Use my attached photo as the product. 0-3s: the bottle in darkness, a single warm light sweeps across the glass and reveals the label. 3-7s: slow orbit around the bottle on a dark stone surface, macro on the dropper. 7-10s: the bottle settles center frame, big type "NOVA" then "Now available", small "Vitamin C Serum" under it. Premium skincare launch, deep black background, warm amber light. Keep the bottle shape and label exactly as in the photo. No people, no small print.

NOVA (a fictional serum) — photo-led launch reveal, 10 seconds, built from the attached photo. Bottle, cap, and label carried over. The end card shows only "Now available", partly hidden behind the bottle; the big "NOVA" type and the small "Vitamin C Serum" line never appeared.

Two things went wrong in this run. First, it took three submissions: the engine's content checker rejected my first two briefs as policy violations — a false positive, nothing in either brief was remotely against policy, and nothing was charged — and the plainer wording above went through. Second, the end card. The brief asked for big "NOVA" type, then "Now available", with a small "Vitamin C Serum" line under it. The clip delivered one of the three: "Now available", with the bottle sitting in front of two of its letters. The small line I asked for (against my own Step 2 advice) is gone, and so is the big wordmark. By the Step 3 standard that's two missing beats plus an occlusion, and the second-take instruction writes itself: "put NOVA in big type above the bottle, then Now available, nothing else." The clip cost $1.60 and took about six minutes to render, against under three for the text-led ones.

Step 5 — Iterate in sentences, not timelines

The second take is where AI launch videos pull away from the traditional process. There is no timeline to reopen. You say what to change — "hold the end card a second longer", "make the accent orange", "remove the sub-captions under the features" for Lumen, "put NOVA in big type above the bottle" for the serum — and regenerate. Because a take costs well under two dollars and a few minutes, you can afford to test several hooks and pick the one that feels like the launch, which is what freelancer pricing never allowed.

Step 6 — Export for where it ships

Generate the format the channel needs rather than cropping later: 16:9 for the website, product page, and YouTube; 9:16 for Reels, TikTok, and Shorts; 1:1 for feeds. Export is MP4 (H.264) — 1080p on Pro, 4K on Ultra — which the App Store, Product Hunt, and the major social networks all accept. On iArt the $2 signup credit covers generating and previewing a full video with no card; exporting the MP4 is a paid feature.

Mistakes that cost the most time

• Adjectives instead of beats. "Exciting, modern, premium" produces a random video. "0–3s countdown, 3–7s name reveal" produces yours.
• Too many words on screen. Three feature beats fit a 15-second teaser; in my runs, more than that stops being readable.
• Physical product without a photo. The engine will invent a product that is almost yours.
• Judging take one on taste. Judge it on compliance first — beats present, nothing extra; taste is what the second take is for.

FAQ

What should I check first when the launch video comes back?

Compliance before taste: every beat from the brief present, in order, with the exact words you asked for — and nothing extra. In my runs the misses were an added sub-caption line and a dropped end-card beat, both of which became one-sentence fixes for the second take.

Can AI make a launch video for a physical product?

Yes, if you attach a product photo. In my run the bottle, cap, and label carried over from the photo into every frame; iArt's own agent warns that the label wordmark is redrawn rather than copied, so keep it large and frontal. Without a photo the engine invents a plausible product that won't match yours.

What does an AI product launch video cost?

On my runs: three text-led launch teasers (15, 20, and 15 seconds) cost $0.86 of credit in total, and a 10-second photo-led product reveal costs about $1.60. Generating and previewing runs on the $2 signup credit; exporting the MP4 is paid.

Do I need a script or storyboard first?

The brief is the storyboard. Writing it as timecoded beats — hook, reveal, proof, CTA — replaces the script, the storyboard, and the shot list in one paragraph, and it is what the engine reads.

What formats should I generate for a launch?

16:9 for the website, product page, and YouTube; 9:16 for Reels, TikTok, and Shorts; 1:1 for feeds. Generate each format from the same brief rather than cropping one master, so the composition is built for the frame.

Make yours

Take the brief that is closest to your launch, swap in your product name, your three beats, and your CTA, and run it on the AI product launch video maker. If it's a physical product, attach the photo. You'll have a first take before the meeting about the launch video would have ended.