Camera to cloud culling: what ships today

Camera to cloud sounds like the answer to a same day deadline, right up until 1,800 frames land in a folder while the client is asking for keynote photos. The transfer got faster. The choosing did not. This is what genuinely ships from a stills camera in 2026, checked against each vendor's own page, and where selection has to sit in a live workflow.

The connection is not the bottleneck

The promise is easy to like. The frame leaves the camera while you are still shooting, an editor somewhere else opens it, the client posts it before the room empties. Nothing about that is fake. What the pitch skips is that moving a file was never the slow part of a live delivery.

The slow part is that somebody has to decide which of the eleven frames of the CEO shaking hands is the one where nobody blinked, the sponsor logo is readable and the hands look right. Camera to cloud delivers that decision to a different room faster. It does not make it. Point a live pipe at an event without an editorial system on the other end and you have built a very fast way to flood a marketing team with 400 near-identical photos.

So read the rest of this in two halves. First, what the hardware and services actually do today, which is narrower and more specific than the marketing suggests. Then where the picking sits, who does it, and what a realistic live promise looks like when you are also the one photographing.

What actually reaches a cloud from a stills camera

Frame.io is the system most people mean by camera to cloud, and its own support page lists the integrations device by device. That list is the useful document, because it separates cameras that send stills from cameras that send video only. Everything in this table is from vendor pages, checked while writing.

Sources, in order: the Frame.io supported device list, Nikon's NX MobileAir help, Canon's image.canon page and Sony's Transfer & Tagging page.

Two things fall out of that table immediately. The first is that if you shoot Canon stills, the Frame.io route does not currently include you: the Canon cameras on the list are cinema bodies sending proxy video. The second is that the phrase "camera to cloud" covers at least three different architectures, one where the camera talks to the service, one where a phone in your pocket does the talking, and one where you are really running an FTP job with a nice interface on it.

The fine print that decides whether it works at an event

Sometimes the phone is the camera to cloud device. On the Nikon route, transfers run through the NX MobileAir app, and Nikon's own help spells out the condition: "Pictures shot while the [Camera to Cloud] dialog is displayed will be automatically transferred to Frame.io." So a phone has to be paired, awake, charged and sitting on that screen for the whole block you are covering. That is a workable setup on a monopod at a conference. It is a different proposition when you are moving through a reception with two bodies.

Free cloud storage is usually short term storage. Canon is refreshingly blunt about it on the image.canon page: uploads are held for 30 days, and "after 30 days original images are deleted unless they have been moved into long term storage." That is fine as a transfer service. It is not a backup, and it should never be the reason a card gets formatted.

Camera to cloud is included, but storage is the meter. Frame.io states that "all Frame.io plans include Camera to Cloud," including the free tier, which as listed on its pricing page gives 2 members, 2 projects and 2 GB of storage. Pro is $15 per member per month and Team is $25, with terabytes attached. For a wedding or a full conference day, storage is the number to plan around, not the feature list.

Prices checked August 2026 from each vendor's own pricing page.

Voice tags and captions are the underrated part. Sony's app supports "hands-free high-speed captioning possible with voice recognition" and IPTC captioning, which exists because press photographers have to caption before they transmit. If you cover events for a client who needs names and titles attached, that is worth more than another megabit.

The files arrive, the decisions do not

Here is the part that catches photographers out. Getting frames into a cloud is not the same as getting them into the place where you cull, and it is definitely not the same as getting your picks back out.

Frame.io connects to Lightroom, and the connection is specific. Per Frame.io's own help page, the integration "only supports images at this time," images uploaded through camera to cloud "will automatically copy C2C images into Lightroom once they upload to Frame.io," the two services are billed and metered as separate storage plans, and there is no return trip: "files will need to be rendered out of Lightroom and uploaded back to Frame.io." Lightroom Classic sits one more step away again, because it pulls from the cloud rather than being the destination.

Read that as a shape rather than a list of gripes. A live pipeline moves pixels one way and leaves you to move decisions the other way by hand. Whatever you flag in the cloud viewer does not automatically become a flag on the RAW file on the card, and whatever you flag at the desk does not become a note in the client's folder. If you want the whole map of what carries and what quietly does not, we wrote it up in syncing picks between devices.

The practical consequence: decide, before the doors open, which copy is the one you cull. Usually it is the local one, because that is where the RAW files live and where your ratings will still exist next year. The cloud copy is for the people waiting.

Two lanes, and a hard cap on the live one

The setup that survives contact with a real event separates a live publishing lane from a complete coverage lane. The live lane is deliberately tiny: only the frames people need while the room is still full. The complete lane holds breakouts, alternates, groupings and everything the final gallery needs. Trying to serve both from one queue is how you deliver late live photos and a rushed gallery.

Make the two destinations before the first guest arrives, named so obviously that a tired assistant cannot confuse them. If your camera has in-camera rating, use one star, and only one star, for live candidates. A detailed rating scheme invented during a keynote is a scheme you will abandon by the second session. There is more on setting that up in in camera rating.

Then sell a live deliverable that a human can actually produce. "Photos during the event" is not a brief. Ten polished frames roughly an hour after doors open, then eight to fifteen after each major block, is a promise you can keep while still photographing. Build each set around roles rather than pretty pictures: a wide that proves the scale, a clean portrait of the speaker, an audience reaction, a branded stage view, one candid interaction. For an awards dinner, add each recipient and a sponsor-visible handoff.

Cap the number and stick to it. At a 500 person gala, 35 varied photos tell a better story than 110 similar ones, and the cap is what forces the choosing. Our full argument on set sizes is in how many sneak peeks to send, and the delivery rhythm around it is in the same day social workflow.

Test the connection before you sell the deadline

Bandwidth is an assignment variable, not a footnote. RAW files are large and venue wifi is shared with several hundred phones, so a live lane built on RAW uploads is a bet you will lose in a ballroom. Send JPEGs for the live lane, keep the RAW files on the cards as the masters, and upload the archive when the urgency has passed. Shooting RAW plus JPEG makes that split trivial, which is one more reason it keeps appearing in rush workflows.

Test from the position you will actually shoot from, at the time of day you will shoot. Send ten files, time it, open them on the recipient's device, and confirm the filenames survived. A test in the empty lobby next to the router tells you nothing about the room after it fills with people. If the venue network is shared, bring your own connection, and find out in advance whether the building has the kind of walls that end that plan.

Treat the cloud copy as an additional copy and nothing more. Cards stay in the wallet until a local ingest exists and the uploaded count matches what you shot. Check that count at every card change, because the failure you are guarding against is not dramatic, it is 40 frames that silently never uploaded during the one speech the client cared about. The rest of the safety net is in the ingest checklist.

What a machine can decide, and what it cannot

AI selection is genuinely good at the grind: soft focus, closed eyes, camera shake, the eleven near-identical frames of one handshake. In an event folder that is most of the volume, and clearing it fast is worth real money on a same day deadline. Adobe now ships Assisted Culling in Lightroom, scoring subject focus, eye focus and eyes open, and the desktop tools have done this for years.

What none of them can do is read the brief. A tool does not know that the CEO's thirty seconds with an investor matters more than a better lit frame of an unknown attendee, or that a half-hidden sponsor logo makes a technically perfect photo useless for the post-event report. It cannot know that the person at the edge of the crowd shot asked not to be photographed. For a speaker, the best frame is often the one with a purposeful gesture and a readable slide, not the widest smile.

So give the machine a defined job: surface the technically sound frames, group the duplicates, put the misfires at the back. Then review the survivors against the shot list and the client's priorities. Editorial importance still belongs to the person who read the brief and stood in the room. Speed only helps if it protects the story. The wider version of that argument, and where each tool sits, is in high volume event culling.

Where Kepla fits, honestly

Kepla is a native app for iPhone, iPad and Mac, and it is still being built. The problem it is aimed at is the one this whole article circles: the connection got fast, the choosing stayed slow, and the choosing happens wherever you are, which at an event is very rarely at a desk. The design is a shortlist, not a purge. Every frame gets checked against clear criteria, sharp focus, open eyes, expression, near-identical frames, and you approve the picks. Nothing is deleted, moved or renamed, so the frames it did not pick are still there when the client asks for the one you both skipped.

The booking half of Kepla is live and free today. If you shoot events, the booking page holds a date and takes a deposit while we finish the picking app, and it costs nothing while we build.

FAQ