This runs in your browser. Your photos are read and measured on this page and none of them are uploaded anywhere.
Drop in the photos you cannot choose between
Or drag them onto this box. A hundred or more is fine, it works through them a few at a time so the page keeps responding.
Reading your photos
Your picks, strongest first
The ones it left out (0), and why
Why choosing is the hard part, not shooting
Taking two hundred photos on a weekend costs nothing. Cutting them to ten is where everyone stalls, for a reason that has nothing to do with taste. You shot eleven frames of the same table because you were checking the screen each time, and at thumbnail size all eleven look fine. Scrolling back and forth between them is the least useful thing you can do with an evening.
Split the job in two. First throw out everything technically not worth posting: the shot where someone moved, the one where the sky is a white sheet, the ten frames that are the same frame. That part is mechanical, and it is what the picker above does. What is left is the interesting decision, which moments you want to tell, and that part stays yours.
The rule that makes a dump feel put together is that no two photos should feel like the same photo. Nine warm indoor shots read as one long blur even when each frame is good. So the picker does not simply keep the top ten by score. Once a strong shot is in the set it holds back anything close to it in colour and framing, and lets a slightly weaker but different photo through instead.
What it measures, and what it cannot judge
Four measurements, taken on a small copy of each photo so a hundred of them do not lock up your browser.
- Focus. A Laplacian filter responds to edges, and the variance of its output is wide for a crisp photo and collapses for a shaky or out of focus one. The number on each card is that variance, judged against the rest of your batch rather than an absolute standard.
- Exposure. The brightness histogram gives the average level, the share of pixels blown to pure white and the share crushed to pure black. Blown highlights are the one thing no editor brings back, so they are penalised hardest.
- Near-duplicates. Each photo gets a 64 bit perceptual hash from a discrete cosine transform of a 32 by 32 grey copy. Two shots of the same scene differ in only a handful of bits even when one is cropped or a shade brighter. Anything inside the threshold is grouped, and only the sharpest, best exposed member survives.
- Colour. A hue histogram weighted by saturation gives the dominant colour of each frame. That drives the variety rule and the colour dot on each card.
Now the honest part. None of this knows what your photos are of. It cannot tell that the soft frame is the only one where your sister is laughing properly, or that the boring photo of a receipt is the joke that makes the set land. It measures whether a photo is technically sound and distinct from its neighbours, not whether it means anything to you. Treat the output as a shortlist, not a verdict.
How to order a dump once it is picked
The picker hands them back strongest first. That is a ranking, not a running order. Reorder your ten along these lines.
Slot one is the whole post. Most people see the first frame and nothing else, so put your best photo there even if it is the obvious one. A face, a clear subject, a colour that stops a thumb.
Alternate wide and close. Two landscapes in a row flatten each other. A room, a hand, a street, a face reads as a story because the distance keeps changing.
Break up the colour. If three golden frames survived, space them out. The colour dot on each card makes that quick to eyeball.
Put the joke second to last. The blurry one, the receipt, the dog under the table.
Close on something quiet. A wide last frame without a face tells people the set is finished. The copy button under the picks writes your filenames out in sequence, so you can select them in that order in your photos app.
How many photos to post, and when to cut deeper
Ten is the default because a carousel holds twenty and about half of that is where attention holds. The practical range is six to twelve. Below six it reads as a single post rather than a dump. Above twelve the tail gets swiped past, and your weakest frames are the ones sitting there.
Adjust with the occasion, not with how many photos you have. One evening out is six to eight: one location, one outfit, and the frames repeat fast. A trip across several days carries twelve, because the light and the setting genuinely change. A birthday sits at eight to ten.
If you are unsure, set the tool to twelve, look at the last three and ask whether you would miss them. That is a far easier question than choosing from two hundred, and the answer is usually no. Then set it to nine and post that.
When to overrule the picker
Every rejected photo shows why it was dropped, with numbers where numbers exist: the hash distance to the shot it duplicates, the share of the frame blown out, how its focus compares to your batch. A tool that says only "not chosen" gives you nothing to argue with.
Overrule it in three cases. When the reason is near-duplicate but the two frames are not the same moment, since expressions change between frames and the hash does not see faces. When the reason is blurry and the blur is subject movement rather than a missed focus, which is often the best frame in a set. And when the reason is just outside, which means nothing was wrong with it at all.
Use the replace button on any pick, then choose a shot from the rejected pile, or drag it across if you prefer. The numbering holds, and the photo you swapped out joins the pile with a note saying you replaced it by hand.
ShutterAI does this curation on the phone, on the roll you just shot, alongside the film looks, Best Face for bursts and captions in your voice. This page is the part we could give away in a browser, and it needs nothing installed.
Questions people ask
How do I pick the best photos for a photo dump?
Cut in two passes. First remove everything technically weak: the blurry frames, the blown out ones and every duplicate of a shot you already have, which is what the tool above does in a few seconds. Then choose from what survives on meaning, and make sure no two of your final picks look like the same photo.
Does this upload my photos anywhere?
No. The picker reads the files with the browser's own image decoder, measures them on a small copy in a canvas, and shows you the result. There is no network request in the tool at all, and closing the tab clears everything.
How does it detect near-duplicate photos?
Each photo is reduced to a 32 by 32 grey square, run through a discrete cosine transform, and turned into a 64 bit perceptual hash. Two photos are grouped when fewer than about twelve of those bits differ, which catches the same scene shot twice even if one is cropped or brighter. You can loosen or tighten that with the grouping control.
Can it tell whether a photo is actually good?
It can tell whether a photo is sharp, well exposed, colourful and different from your other photos. It cannot tell whether it matters to you, whether the expression is right, or whether the framing is a deliberate choice. Treat the picks as a shortlist and overrule anything that deserves it.
Will it work with a hundred photos or more from my camera roll?
Yes. It processes them one at a time with a progress bar and hands control back to the browser between each one, so the page stays usable while it works. Very large files simply take a little longer to decode. It stops at 400 photos in one go.