京都駅から清水寺、伏見稲荷、祇園、東福寺、嵐山へ。紅葉と雪の夕暮れを、歩いて、空からFrom Kyoto Station to Kiyomizu-dera, Fushimi Inari, Gion, Tōfuku-ji and Arashiyama: an autumn evening and a snowy one, on foot and from the air
About 5.8 by 8.6 km of Kyoto around the station, from Fushimi Inari to Ginkaku-ji, plus Kinkaku-ji and Arashiyama–Sagano, rebuilt in 3D to walk or fly over: through the thousand torii, past the thousand Kannon of Sanjūsangen-dō with its 28 guardians and the gods of wind and thunder, out on the Kiyomizu stage and Tōfuku-ji's Tsūten-kyō above valleys of maples, to Tō-ji's pagoda, Gion, Eikandō, the bamboo grove and the Togetsu-kyō. An autumn evening and a snowy one (15 cm). The city is generated from PLATEAU (MLIT), OpenStreetMap and GSI elevation data; the landmarks are modelled by hand in Blender. Light baked in Blender, drawn with WebGPU. Bach's Goldberg Variations, played slowly by a piano synthesised in the browser. Japanese and English. Needs a WebGPU browser; the first load is about 160 MB (the rest streams in as you go).
The whole main island of Venice rebuilt in 3D, to walk, to ride by gondola along three canal routes with a rowing gondolier, or to fly over: Piazza San Marco and the Basilica (the mosaics of its west front, five domes inside), the Campanile, the Doge’s Palace and the Bridge of Sighs, the Rialto, Caffè Florian and La Fenice. Buildings are generated from OpenStreetMap, with the landmarks modelled by hand; light is baked in Blender and drawn with WebGPU. Solo piano, water, bells and pigeons are all synthesised. Japanese and English. Needs a WebGPU browser; the first load is about 100 MB (the rest streams in as you go).
Vatican City rebuilt in 3D and toured in a nine-minute drone flight: Bernini’s colonnade and the obelisk, Michelangelo’s dome inside and out, the Pietà and the baldachin, the Momo staircase, the Laocoön, the Gallery of Maps, the Raphael Rooms, and the ceiling and Last Judgment of the Sistine Chapel. Buildings from measured drawings and OpenStreetMap, light baked in Blender, artworks from public-domain images and 3D scans. Free flight and a place menu; synthesised piano with the fountains and bells. Japanese and English.
A 3D reconstruction aiming to show the basilica on 1 October 2026, based on official reports and photographs verified through 22 September 2026. Walk and fly among branching columns and stained-glass light, the Nativity and Passion façades, and the unfinished Glory façade. Desktop WebGL 2; initial download about 325 MB. Details, sculptural depth and lighting include interpretation. Sources and licenses are included.
A dome-screen flight ride, in a browser, full screen. From the clouds at dawn over an ice spire, great falls, a coral atoll, a red canyon and an aurora over the ice, to a night harbor where you fly through the fireworks. Every landscape is ray-marched live on the GPU; the solo piano score, which changes with each place, and the sound of the water are synthesised. Japanese and English.
A 3D study of Notre-Dame before the April 2019 fire, reconstructed from photographs and plans. Walk through the nave and chapels, press F to fly and E/Q to rise or descend for a closer look at the roses and vaults. For desktop WebGL 2 browsers; the initial download is about 258 MB. This is an interpretive reconstruction, not an official survey model. Full source credits, modifications and licenses are included in the Credits page.
A trench run drawn only in glowing vector lines. Thread the gaps in the bulkheads, shoot down turret fire and fighters, and put the last torpedo into the exhaust port. Mouse, keyboard or touch. Japanese and English.
テーマを書くと、この PC が動画をつくるWrite a theme and my PC makes a film
テーマと長さ（2〜3分）を書くと、絵も声も音楽もそろった、二人の掛け合いの解説動画を、私の PC が10〜15分でつくります。台本と挿絵は Claude、声は VOICEVOX、書き出しは HyperFrames。できあがった動画はその場で見られ、チャプターから飛べます。サンプルは「忘れない復習のタイミング」：180.000 秒ちょうど、声の重なり 0、自動検査 18 項目すべて合格。合言葉を知っている人だけが使えます。
Write a theme and a length (two or three minutes) and my own PC makes a two-voice explainer film in ten to fifteen minutes — pictures, voices and music included. Claude writes the script and draws the illustrations, VOICEVOX speaks, HyperFrames renders. The film plays right there, with chapters to jump to. The sample runs exactly 180.000 s with no overlapping voices and 18 of 18 automated checks passed. Only people who know the passphrase can ask for one.
A closed sketchbook under a warm desk lamp. Open it and ten pages draw themselves, stroke by stroke, in a handwriting I invented: the questions I keep circling, one page torn out and taped back, and a blank last page for you. The felt-piano score and the pencil sounds are made in code too.
The sketchbook of Juno, a fictional illustrator. Dot-grid pages peel over like real paper and every drawing pencils itself in: six works, process notes, a try-it page that draws what you ask for, and a letter in an airmail envelope.
A headline split into RGB, eight works generated live in the browser, and a biography that decodes itself out of noise. Now and then the screen tears; one Stabilize switch calms it all. Japanese and English.
A fictional port town, 1947. A torch singer's only alibi is a lipstick-marked glass. Five scenes, three endings, and the only colour on screen is the crimson of the one clue that matters.
