A time machine layer over every place on Earth. Point your phone at a street and see it move in 1900, 1950 and 2080, built with local archives and the people who lived there, and always honest about what is record and what is reconstruction.
2. Why people can't resist it
Standing in two times at once. The street under your feet, moving in 1950, is a feeling nothing else gives you.
Your street, your grandparents. Every place has someone's story. Locals adding "I bought bread here in 1978" make it personal and addictive.
The future is a choice. Toggling the same corner between greening and flooding makes climate concrete, and shareable.
3. How it works for real
Locate: GPS plus a vision model (Gemini or GPT vision via OpenRouter) matches the photo to the street and reads its current features: building heights, road layout, trees, signage.
Ground: retrieval over open archives (city photo collections, old maps, census and planning data, Wikimedia Commons with licences) for the exact block and decade. Each claim carries a source or the label "reconstruction".
Describe: a frontier LLM (Claude or GPT) writes a per-era scene spec: what vehicles, clothes, signage, building heights and trees, plus a 2 sentence narrated caption.
Render: an image model (Flux, or Gemini image) redraws the photo per era with structure guidance so the street geometry stays fixed. A video model (Wan 2.x or Seedance; Veo 3.1 Lite for premium) animates 4 seconds of street life. Free tier uses the still with parallax motion.
Future: 2050 and 2080 scenes take sea level and heat projections from public climate datasets for that coordinate, in two scenarios.
Voice: TTS narration in the local language and English.
4. Cost per use
Vision match and scene specs: about $0.02. Six era stills: about $0.15. Animated 4 second clips at $0.10 to $0.50 per second: $0.40 to $2 each, so about $2.40 to $12 for all six eras.
Free: 3 windows a day with stills and parallax motion. Traveller ($5 a month): animated clips for every era, offline city packs. Cities and museums: license curated windows for walking tours, paid per scan via x402 so a tourism app can embed any window by request.
5. Growth loop
The before/after slider card is made for sharing: "my street in 1950 vs 2080". Every window asks locals for a memory, and each contributor gets notified when someone opens their corner. Schools run "map your street" projects that fill a whole neighbourhood. Every Sunday a new window opens near home, unlocking more eras on your neighbourhood map, and a feed shows when locals add memories to places you have opened.
6. Risks and guardrails
Honesty: every scene shows "AI reconstruction" and links the archive records used. No fake "historical photos" exported without that label burned in.
People: crowds are generic silhouettes. No real faces of living people are generated; uploaded photos are processed and not kept unless you save them.
Contested history: colonial, wartime and displacement history is described with sources and community review, never erased or sanitised.
Memories: contributions are moderated, can be anonymous, and the contributor can remove them anytime.
7. MVP build order
Photo plus GPS to place, with a vision model reading the scene.
Era slider with image-model redraws for 3 eras, labelled reconstructions.
Archive retrieval for one pilot city (open photo archive), facts card with sources.
Local memories with moderation, and the before/after share card.
Future scenarios from climate data, then animated clips.