Industries / Retail, hospitality and personal services / Hospitality and tourism / Wedding and events
Photography and video
This work records weddings, conferences and other events in photographs and film, from posed portraits and candid shots to aerial footage, edited films and live streams for people who cannot attend.
5
Milestones achieved, from 1854 to 2020.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1854
Royal wedding restaged for camera
Queen Victoria and Prince Albert married on 10 February 1840, and in 1854 they "re-enacted their wedding for this photograph", by the Royal Family's own account. Early that year they had commissioned Roger Fenton to photograph the family, and on their 14th wedding anniversary he photographed a play their children performed for them.
LessonThe wedding photograph began as a portrait posed for the camera, not a record of the day.
Credit
Queen Victoria and Prince Albert; Roger Fenton, whom the Royal Collection Trust credits with the 1854 royal portraits
2006
Conference talks go online
TED posted its first talks online, free, on 27 June 2006. By September they had passed 1 million views, and in 2007 TED rebuilt its website around them, by its own account.
Credit
TED
2008
Stills cameras shoot HD video
Canon's EOS 5D Mark II recorded full high-definition video on a stills camera; Canon calls it the industry's first SLR to do so. Canon's museum says broadcasters noticed it "facilitated the integration of video and still images", and it won the TIPA, Camera Grand Prix and EISA awards in 2009.
Credit
Canon
2016
Rules for drone photographers
The US Federal Aviation Administration's Part 107 rule took effect on 29 August 2016 and barred small drones from flying over people not taking part in the flight, which includes wedding guests. The FAA wrote that a drone used for aerial photography faces "the same operating restrictions" as one used for bridge inspection. A later rule, published on 15 January 2021 and in force from spring 2021, allows some flights over people.
Credit
US Federal Aviation Administration
2020
Weddings by video call
New York's Executive Order 202.20 dropped the need to apply for a marriage license in person, and New York City's Project Cupid moved licensing online for the first time in the city's history, going live in May 2020; couples could also marry over a video call. Video ceremonies ended on 25 June 2021, while online license appointments are still offered in 2026.
LessonThe online license stayed, but the ceremony went back to being held in person.
Credit
Governor Andrew Cuomo's executive order; the New York City Clerk; the New York City Council
Watch for
Culling and editing while uploading
Imagen says its Intelligent Studio, rolling out to existing customers from 29 September 2026, picks the best shots of a wedding and edits them while the rest of the photos are still uploading, and has an assistant that carries out multi-step edits from plain-language instructions. The article reporting it is marked sponsored. A rival, Aftershoot, claims "100% AI precision" on its home page with no published measurement.
Achieved whenA test not run or paid for by a tool maker measures how often the picks and edits match the photographer's on real weddings, and how much time they save.
Claimed by
Imagen, September 2026
Bystanders ask cameras to blur
Researchers at the University of California, Irvine, built BlindSpot, a phone system that lets a bystander ask nearby cameras to blur their face with a swipe of the hand, a blinking light or a radio tag. They report near-perfect recognition of the gesture at 1 to 2 meters, about 90% accuracy for the light, often over 95% for the tag, and weaker results with more than eight people in view.
Achieved whenAn outside group repeats the tests, or a camera or event-photo product ships the feature and publishes its miss rate.
Claimed by
University of California, Irvine, December 2025
Open
01
Blur guests who opt out
Software that finds every guest wearing a conference's "no photos" lanyard in the event's pictures and blurs their face before the pictures are published, with a measured miss rate. The test set is photos from an event, or a staged one, where every person photographed has agreed to take part and each person's lanyard color is recorded.
The briefCredit · unclaimed02
Culling scored against photographers
A public test of AI culling tools, which pick the keepers from the thousands of frames shot at a wedding. Collect complete, unedited shoots from photographers who agree, with the couples' consent, together with the set each photographer delivered. Run each tool on each shoot and score how many of the photographer's picks it keeps and how many of its picks the photographer rejected. Publish the scores, not the photos.
The briefCredit · unclaimed
Open 01 · the brief
Blur guests who opt out
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
Software that finds every guest wearing a conference's "no photos" lanyard in the event's pictures and blurs their face before the pictures are published, with a measured miss rate. The test set is photos from an event, or a staged one, where every person photographed has agreed to take part and each person's lanyard color is recorded.
Why it is still open
Conferences such as PyCon UK 2017, DjangoCon Europe 2018 and GUADEC 2022 hand out colored lanyards so people can opt out of photos, and leave photographers to spot them; GUADEC's policy tells photographers not to photograph people in red lanyards and to delete pictures on request. Berkeley's Respectful Cameras (2007) blurred construction workers who wore green vests and aimed to miss under 2% of faces, and UC Irvine's BlindSpot (2025) responds to gestures, lights and radio tags; neither was tested on event photos with lanyards. This research found no tool that blurs by lanyard.
alexwlchan.netevents.gnome.orggoldberg.berkeley.eduhelpnetsecurity.com
Built with what exists
Open models that find objects from a text description, such as Grounding DINO or YOLO-World, to find lanyards; a face detector to link each lanyard to a face; a segmentation model such as SAM 2 for clean blurs; Python and OpenCV.
First step
At one meetup this month, with everyone's consent, take 200 photos in which some volunteers wear red lanyards, and label every face by hand.
Done looks like
On at least 2,000 photos with at least 100 opted-out volunteers, a published share of opted-out faces left unblurred, measured against the 2% that Respectful Cameras aimed for, the share of other faces blurred by mistake, and code someone else can rerun.
Open 02 · the brief
Culling scored against photographers
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of AI culling tools, which pick the keepers from the thousands of frames shot at a wedding. Collect complete, unedited shoots from photographers who agree, with the couples' consent, together with the set each photographer delivered. Run each tool on each shoot and score how many of the photographer's picks it keeps and how many of its picks the photographer rejected. Publish the scores, not the photos.
Why it is still open
The only accuracy figures found come from FilterPixel, a culling company that tested its rivals: it reports 91.2% keeper identification for Aftershoot and 63.4% accuracy for Imagen, against picks by three unnamed photographers, and the same page gives two definitions of accuracy and image counts that do not agree. The Princeton photo-triage benchmark (2016) has 15,545 photos in 5,953 short series from personal albums, not whole events. This research found no independent test on full wedding shoots.
Built with what exists
Trial versions of Aftershoot, Imagen, FilterPixel and Narrative Select; the Princeton benchmark's method for scoring choices within a burst of near-identical frames; a spreadsheet of each photographer's delivered set.
First step
Ask two wedding photographers for one past shoot each with their delivered selection, and run one tool on both.
Done looks like
At least 20 full shoots from at least 10 photographers, scored on at least four tools, with a public table of the share of delivered photos each tool kept and the share of its picks the photographer rejected, plus a second photographer's selection on five shoots to show how far two people agree.