Start with one question
A video that tries to cover a whole country rarely holds attention. A video that answers one question does. Write the question in a sentence, then ask for a script of about 100 to 130 words, which is 45 to 55 seconds of speech.
Open with the hook, which is the surprising part, not the background. The map should show the answer while the voice explains it.
Let the agent build the scenes
Ask your connected AI for the video and say the format, usually vertical. It writes the script, records narration with your Gemini key, times the captions with your Groq key and saves a scene for each place, timed to the words.
Ask for what you want to see, such as a stat card for a number, a flag when a country is named, or an arrow showing a distance. The more concrete the request, the better the result.
Check it before you export
Open the preview link and watch it through once with the sound on.
- ›Does each outline appear as its place is named?
- ›Is every important label readable on a phone screen, clear of the captions?
- ›Does the camera stay on the place long enough to read it?
- ›Does the last second invite the viewer to follow or watch the next one?
Fix and export
Change anything by hand in the editor, or ask your AI to change it, for example "make the camera look straight down in scene 3". Then export. Exporting runs in your browser, needs a Pro or Studio plan and your free Mapbox public token, and produces an MP4 ready to upload.