Rushes: A living wall for pics

Point it at a folder and your whole collection comes up at once as a single surface you can fly through; every clip and motion photo playing in place, nothing streamed, nothing uploaded, nothing indexed by anyone else.

Rushes is the film term for the day's raw unedited footage, viewed all together.

Rushes turns a five-year camera roll into one navigable, moving wall you can
search by colour, by subject, or by person -> entirely on your own machine.

In Short:
Every photo app shows you a grid of stills and asks you to remember when
something happened. Rushes shows you everything you own in one continuous space and lets you find things by what they actually look like.

- The whole library, one surface: thousands of items in a single instanced draw call. Zoom, pan and spin at frame rate on ordinary hardware.
- Everything that moves, moves: motion photos and clips play in place, all of them at once, at effectively zero marginal cost. Brush the wall to wake motion under your cursor, or let the whole thing breathe.
- It plays the photos that are secretly video: weld a full MP4 onto the tail of a JPEG. On a typical roll that's ~12% of your photos carrying hundreds of megabytes of footage no other viewer on your machine has ever played. Rushes plays it : and exports it as a real MP4, byte-exact, no re-encode.
- 8 ways to see the same library: Real clock time, capture order, hour of day across five years, hue and colourfulness, one spiral turn per year, motion vs. stills, a day-of-year calendar, and a layout where position *is* subject. Press a key and the library flows to the new arrangement; hold a key to lean between 2.
-Search by what's in the frame: teal finds pictures that genuinely contain teal, not ones whose average is muddy brown. `a dog on a beach at sunset` works with no tags and no label list. Name a face once and that person becomes a filter.
- Nothing leaves the machine: all models run locally. No account, no cloud, no upload. The tagging pass is optional at every level — remove the models and the app still starts and works.
- It records itself: One key renders a scripted tour of your library straight to a deterministic MP4, or writes a contact-sheet PNG.

Dou you want years of phone photos, family video, project footage? , want to *see* it rather than scroll it?: photographers and editors triaging rushes, anyone digging a specific memory out of five years of undifferentiated grid, and anyone who refuses to hand a personal library to a cloud service to get search.

What makes it different? - Typical gallery app vs Rushes :
- Paged grid of stills vs The entire library as one zoomable, rotatable wall
- Video is a thumbnail you tap vs All motion plays simultaneously, in place
- Motion-photo video is invisible vs Detected, played, and exportable losslessly
- Sort by date, maybe album vs Eight designed layouts with a continuous morph between them
- Search by filename or cloud tags vs Colour histograms, natural-language meaning search, local face grouping|
- Library lives on someone's server vs Library and models live on your disk
- Layout reflows when you resize vs Position never changes, so spatial memory survives
- Filtering re-sorts the grid vs Filtering desaturates in place; nothing moves

Some limits:
- Rushes reports rankings, not verdicts ; meaning search tells you what is *most* like your phrase, across everything.
- Face grouping is empirical and will get somebody wrong, so merging, splitting and naming are first-class controls, not recovery paths.