SnapPipe

In news and event photography, time defines the value of a photo: the first photo online gets the most views. Today a photographer must pull the SD card, copy it to a laptop, and sort thousands of near-identical frames before posting, and miss the next moment while doing it. Freelancers have no photo editor to do this for them. SnapPipe moves that work to the moment of capture. A Raspberry Pi Zero 2 W on the camera cage reads each shot over USB (gphoto2) and uploads it to a self-hosted server, resumable and newest first, so nothing is lost when Wi-Fi drops. The server sends photos to Meta Muse Spark 1.3. A whole burst is sent in one request so the model compares frames with each other and picks the best one. It returns scores, star ratings, tags and issue flags as schema-validated JSON. Ratings and tags are written into the photo’s own XMP/IPTC metadata, the file is backed up to a NAS, and a live web album sorts everything by stars and tags. The photographer opens it on a phone, downloads the pick and posts, or shares one link with the client. For sports, a team roster can be added: the model reads jersey numbers and drafts a caption, but it may only refer to players as placeholders ({P0}). Names come from the roster in code, and low-confidence matches are flagged for human review. There is no face recognition.