How to make a product launch video with AI: a Kuchi walkthrough
A good launch video is mostly timing: when each shot starts, how it enters, how much of the frame moves and when. Kuchi gives your AI agent that timing, measured frame by frame from real launch videos, and your agent builds a video with the same rhythm and your product in every shot.
Here is what the whole process looks like, from the first prompt to the rendered file.
What you need
- An AI coding agent that can run code and render video, such as Claude Code, Codex or Cursor, with the Kuchi MCP server connected. Chat apps like Claude and ChatGPT can do the first three steps and hand the plan to a coding agent.
- Your own material: your logo (SVG or transparent PNG), screenshots or screen recordings of the product, the one to three features to show, and the closing call to action.
- Node.js, so the agent can create a Remotion project and render.
Step 1: Choose a template
Tell your agent about the product and the feel you want:
Make a 30-second launch video for my app, a calendar for remote teams. Clean and confident, light theme. I have a logo and five screenshots.The agent calls kuchi_suggest_templates and shows you a short numbered list of real launch videos, each with its poster and link, what fits your product, what differs and what you would need to supply. Choose by number, or open the launch video library, pick any video and paste its link.
Step 2: Study it shot by shot
kuchi_get_breakdown returns the measured structure of the video you chose: every shot's exact start and end frame, whether it enters on a cut, a transition or a hold, how much of the picture moves, the energy curve second by second, and the colour roles (background, secondary, accent). It also returns storyboard sheets of real frames.
Kuchi does not say what a shot shows. Your agent looks at the frames and writes one line per shot: its kind (title card, kinetic type, UI screen, device, logo and so on) and what is on screen and where.
Step 3: Plan with your material
The agent sends your product details and its shot analysis to kuchi_plan_remake. If anything is missing, the plan says exactly what, worded for that template (for example how many screenshots it needs, and at what size), and the agent asks you in one message.
When nothing is missing, the plan holds a blueprint: each reference shot re-timed to your length on whole frames, with your content assigned to it. A 4.1-second shot in a 60-second reference becomes about 2 seconds in your 30-second cut, on an exact frame, never rounded to a tidy half second.
Step 4: Build it
The agent writes the video in Remotion one shot at a time: the same start and end frame, the same entrance, the same amount of motion and the same layout as the reference, with your logo, screens and words. Your brand colours take the reference's colour roles.
Where a shot fits a ready-made part, the agent installs one of Kuchi's 178 video components instead of writing it from scratch. How that works is in Remotion components your AI agent can install.
Step 5: Check against the reference
Before delivering, the agent runs kuchi_check_remake on its shot list. It flags anything taken from the reference (brand words, footage, UI), any invented logo or metric, any placeholder, and any shot that runs long or short against the shot it rebuilds. It also returns pairs of times that show the same moment in both videos, so the agent can export its frames and compare them side by side, then fix what differs and render the MP4.
What is copied, and what is yours
| Copied from the reference | Replaced with your own |
|---|---|
| Shot timing and order | Logo and product name |
| How each shot enters (cut, transition, hold) | Product UI, screenshots and recordings |
| How much of the frame moves, and when | Words on screen, features and claims |
| Layout, type hierarchy, camera moves | Photos, illustrations and characters |
| Colour roles: background, secondary, accent | The colours themselves, and the music |
Kuchi never puts a frame of the reference in your video and never invents your logo, UI, features or numbers. A missing asset becomes a clearly marked placeholder, and your agent tells you about it.
What it looks like in practice: our own launch film
We made Kuchi's launch film the same way: a Remotion project at 1920×1080 and 30 frames a second, built from Kuchi registry components and nothing else on screen. Some of what we learned:
- One shot list. Every shot lives in one array with its frame count, so a retime is a one-number change and the transitions move with it.
- One look. A white ground, one ink colour and a single red accent, passed to every component through one theme constant.
- Measured, not eyeballed. We compared every rendered frame with the one before to find stretches where nothing moved and cuts that went from a frozen frame to a frozen frame, then gave those shots more or less time.
- Screen time over speed. The first cut felt rushed. Slowing the busiest shots and holding titles until they had finished typing made the film longer and better.
- A transition as a component. Where a full-screen picture cut hard to a white page, a halftone Dot Bloom carries one into the other instead.
Common mistakes
- Rounding shots to half seconds. The rhythm lives in the exact frames.
- Copying the reference's words or UI. Only its motion and taste carry over.
- Adding camera shake, extra shots or new transitions the reference does not have.
- Skipping the check. Comparing frames side by side catches layout drift that a quick watch misses.
Questions
Can AI make a product launch video?
Yes. With Kuchi connected, a coding agent such as Claude Code or Codex picks a real launch video as the template, plans your version to the frame and builds and renders it in Remotion with your logo, screens and copy.
Does Kuchi render the video?
No. Kuchi measures and plans; your agent builds and renders the video with your own tools, so the file and its source stay with you.
What does Kuchi cost?
Nothing. Kuchi is free: the MCP server, the launch video library and the components need no account or API key.
Can I make a vertical video for TikTok or Reels?
Yes. Choose a vertical template, or tell the agent the aspect ratio you want when it plans the remake.
Can ChatGPT make the video?
ChatGPT can choose the template, study it and write the frame-timed plan. A coding agent then builds and renders it.
Will my video copy someone else's launch?
Only its timing, transitions, layout and colour balance. Kuchi's check flags any brand word, footage or UI taken from the reference.

