Inspiration
Standing under Hope, the 25-metre blue whale skeleton in the Natural History Museum's Hintze Hall, we read her label: a few lines of text for an animal that was stranded off the Irish coast in 1891, sold to the museum for £250, and renamed in 2017 as a symbol of hope. It's an incredible story, and almost nobody hears it. Kids glance at the label and walk on.
Museums are full of objects with extraordinary lives, frozen behind glass and a wall of small print. We wanted to give every exhibit a voice.
What it does
Point your phone at an exhibit and take a photo. A couple of minutes later, museum_alive gives you a short animated film in which the exhibit tells its own story, with a consistent character, voice-over, music, sound design and subtitles. You can switch the style (for kids, for experts) or the language.
How we built it
Museum Alive is an agentic pipeline:
- Understand: a vision model identifies the exhibit and reads the museum label, which is our primary source of truth.
- Ground: the agent gathers facts from trusted sources and keeps the links, so nothing in the script is invented.
- Storyboard: an LLM writes a shot-by-shot script as structured JSON: narration, camera, action and mood for each shot.
- Animate: each shot is generated in parallel on fal with Vidu reference-to-video, using our original character sheet for Hope so she looks the same in every scene. We used Seedance 2.0 for hero shots in our demo.
- Assemble: audio is the master timeline. Each clip is automatically trimmed, slowed or held to match its narration, then stitched with
Built With
- agent
- claude
- fal
- seeddance
