Everyone has them: the shot that looked fine on the small preview and turned out completely soft when you actually looked closely, or the one you kept "just in case" the sharper version didn't turn out. They rarely get deleted, because finding them means scrolling through everything else first.
Why blurry photos survive so long
Nobody deliberately keeps a blurry photo — it just never surfaces as a decision. It's not in an album, it's not tagged, it's just one thumbnail among thirty thousand others, indistinguishable at a glance unless you happen to scroll past it and stop to look.
What actually makes a photo "blurry" in a way software can detect
Sharpness analysis works by measuring edge contrast — in a sharp photo, edges (the boundary of a face, the line of a building) have a sudden, high-contrast transition between pixels. In a blurry photo, that transition is gradual and soft. This is measurable mathematically, frame by frame, without any judgment call about content — it's a property of the pixels themselves, which is exactly why it can be automated reliably.
A manual approach, if you want to spot-check first
Switch Photos to Days view rather than the default grid — blur is much easier to notice at a slightly larger thumbnail size than in the tightly-packed default grid. Focus especially on: photos taken in low light (motion blur from longer shutter times is common), photos of moving subjects (kids, pets, sports), and anything taken while walking.
The case for an automatic scan
Manually finding blur means opening photos you'd otherwise never revisit — the exact opposite of an efficient cleanup. An on-device sharpness scan can process an entire library and surface only the genuinely blurry results, which is the category Arc is built to catch: on-device analysis, only your yes required before anything is removed.
Slight softness isn't the same as blur — a shallow depth-of-field portrait with a sharp subject and a soft background is intentional, not a mistake. Good blur detection looks at focus on the primary subject, not overall image softness.
Let blur detection find candidates, not make the verdict
A blur score can spot missed focus and camera shake, but it cannot know whether motion, atmosphere, or the only photo of a moment makes an image worth keeping.
- Review groups, not a giant grid. Compare blurry candidates with nearby shots from the same minute. A sharper replacement makes deletion easier.
- Zoom into the subject. A soft background is intentional in portrait mode; a soft face usually is not. Judge the part the photo is actually about.
- Protect unique moments. Keep the only image of a person, pet, document, or event even when it is imperfect. Technical quality is not the only value.
- Start with severe misses. Black frames, pocket photos, accidental floor shots, and completely unfocused images are the safest first batch.
- Use a small threshold. Delete 20, inspect Recently Deleted, and adjust. Do not approve thousands based on one confidence label.
A useful decision rule
Automatic scoring should reduce scrolling. The final choice belongs to the person who understands the subject and the memory.
Before you finish
- A sharper alternate exists where expected.
- Faces and text were checked at full size.
- Unique moments were excluded from bulk deletion.
A calmer next step
If a blurry photo has a story, keep it. If it is only a failed version of a better shot, you can let it go without guilt.