Walk demo

Walk through Ginza for seven seconds.

This short demo is built from real michiyomi API data. It moves through a side street from a pedestrian viewpoint and shows how the same place appears as structured, queryable language data.

The video has audio. Unmute the player to hear the short conversation; English captions are on by default. Ginza, a 6.7-second loop. Imagery adapted from Mapillary contributors (CC BY-SA 4.0). In-video mini-map: GSI Tiles.

00:00–00:04“Isn't the sidewalk around here a little narrow?”
00:04–00:05“Yeah, it is.”
Read this place

Keep reading
this place.

The walk takes place on a side street in Ginza 1-chome, Chuo City. Continue with the neighborhood profile or the map.

Ginza 1-chome profileA written profile of the whole neighborhood blockOpen on the map → The photo in the videoCaptured in 2022. The values in the video come from this recordOpen on the map → Find similar streetsSearch by words, such as narrow side streets (search runs in Japanese)Find scenery →
How to watch

How to read
the screen

The left half is an instrument panel assembled from michiyomi API responses. The right is street imagery from a pedestrian viewpoint, with the current position at the upper right.

Left 1

Nearby context

describe_location first reports how much data exists near the coordinate, then lists observations for the nearest scenes. Zero results means not recorded, not nothing there.

Left 2

Evidence image

The actual street image (Mapillary) behind the text. Each observation carries a scene ID and capture year so it can be traced back to its source.

Left 3

Only what is visible

A description based only on what the image shows. Values such as a 0.7 m walkable width are image-derived estimates with confidence ratings.

Left 4

Choosing what to say

How one statement is chosen from the evidence. Conditions belong to the capture year and are not asserted as present-day facts.

Upper right

Mini-map

The simulated walking position (map: GSI Tiles).

Behind the demo

What runs
behind it

The statements in the video are based on the production /v1 API. No registration or API key is required.

Step 1A coordinate

The latitude and longitude of the walking position. That is the only input.

Step 2How much evidenceGET /v1/coverage

Before answering, it reports how much data exists for the place.

Step 3Words with a capture yearGET /v1/scenes/nearby

Observations of nearby photos, each with its capture year and the AI generation that read it.

Hear it now

Hear this place
in English.

Fetch the production record for the scene in the video. The page shows a human-reviewed English rendering, keeps the authoritative Japanese observation below it, and reads the English with your device's voice.

GET /v1/scenes/1152690868598943

Press the button to fetch the production API record.

Observations describe the capture year (the scene in the video was captured in 2022), not present-day conditions. Widths and similar values are image-derived estimates with confidence ratings. This demo does not decide whether a place is safe or dangerous; use it to shortlist places to check on site.