iPhone · Clean · Crop · Proof

There's a good photo in there somewhere.

State-of-the-art object removal, AI crop selection, and AI photo enhancement.

Wide shot of Sam's Town Point: band on stage and a woman watching from a table.

Black-and-white crop of the same frame: the woman at the table watching the guitarist.

As shot

Found inside

Download on the App Store App Store

Step 1·Clean

Eight removals on one photo.

01 · The frame

A food truck parked on a corner lot at dusk. Power lines cross the whole sky, a utility pole stands at the left, a traffic cone and cables sit on the ground, and distant cars and a gas station sign clutter the background. — As shot · 1195 × 896

02 · What it found

The app scans the frame and returns labels. Eight recommended here, three flagged and left alone.

overhead power lines distant utility pole distant gas station sign distant cars orange traffic cone white arrow painted on ground cables on ground behind trailer concrete block under trailer wheel wooden utility pole yellow bollard black metal fence

You pick. Anything it missed, you type.

03 · The mask for one label

Take overhead power lines. The app locates that object and paints it into a mask, with a second pass for wires down to the roofline. The mask is the whole permission the AI gets.

The mask laid over the photograph in amber, covering the sky and the wires while the truck, the forecourt and the buildings stay dark. — The mask over the frame it belongs to

The raw mask: white where the power lines and the sky around them were located, black everywhere else. — mask.png · 32.96% of the frame

The ownership map over the photograph: a tighter amber region than the mask, following the wires and the sky the compositor measured as changed. — ownership.png · measured against the render, then feathered

The photograph with the overhead power lines gone. The truck, its Japanese lettering, the OPEN sign, the graffiti and the forecourt are unchanged. — The plate. Wires gone, rest of the frame as shot

The render is compared against your original on colour, edges and structure. Ownership grows from the located object through pixels that changed, then feathers. Repaint too much and the render is rejected and rerun.

Inside the mask

10.85

mean levels changed, 0–255
Outside the mask

0.048

mean levels changed, which is the JPEG this page serves
Before encoding

0.0000072

what the run recorded outside ownership

04 · Eight passes, one mask each

The food truck photograph as shot, with power lines across the sky, a traffic cone, ground cables and a painted arrow. — As shot

The same photograph after eight removals: the sky is clear, the cone, cables, arrow, distant cars and distant sign are gone, and the truck is untouched. — As shot

Drag the line ↔

Truck lettering, menu board, graffiti and the pole marked keep survive all eight.

05 · The OPEN sign

Small lettering doesn't survive a generative round trip. Your device reads the text in the original, checks it against the render, and pastes the camera's pixels back where the lettering changed.

Close crop of the food truck window: a red and white sign clearly reading OPEN, under a large painted Japanese character. — As shot

The same crop straight from the image model: the sign now reads garbled letters, and the painted character has lost its strokes. — Model output

The same crop as shipped: the sign reads OPEN again and the painted character is intact, because the camera's own pixels were put back. — Shipped

Name a sign as the thing to remove and it stays gone.

Step 2·Crop

Twelve crops, judged down to one.

01 · The field

Geometry cuts the field: regions expand into iPhone aspect ratios only, never freeform. Any candidate that splits a face is dropped before voting.

02 · The bracket

Three rounds · twelve judges at every stage

Twelve artists rank their three best each round. First place carries the most weight. The winner reports why it won. You can override it, and every candidate stays available.

03 · The ballot

Candidates go out as one numbered sheet and ballots come back as numbers, so a judge ranks framings without knowing which one the app likes. One judge failing doesn't stall the round.

Step 3·Proof

One frame, several looks.

Regrades that change light, colour and grain — not the things in the frame.

The food truck as shot: even daylight, blue sky, black markings on a white trailer. — As shot

The truck with a different colour grade — Look 1

Light, contrast, colour, density and grain move. The truck, wires, menu board and painted characters hold.

A second scene

A mechanic's shop at night, as shot: a pale blue van under a lit sign reading Quality repairs by expert mechanics since 1990. — As shot

The same night shop with a different colour grade — Look 1

Read the sign in every look: Quality repairs by expert mechanics since 1990.

Full size

Proofs render small. For a full-size file the app measures the colour difference between your photo and the approved proof, builds a colour table, and applies it to your full-resolution original. You export your own pixels, regraded.

Runs on photos already in your library. iPhone, iOS 17. No account, no ads.

Download on the App StoreApp Store