Video to Long Exposure

How to make a long exposure from a video

1. Open the video, or Add photos (many at once: a burst, or interval shots of the night sky). Photos are stacked in file-name order, so IMG_2 comes before IMG_10.
2. Choose the clip. Drag the two handles on the frame strip or type the start and end. It starts as the whole video, or the first 300 frames of a longer one.
3. Choose which frames to use. Every frame gives the smoothest trails. Every 2nd or Every 4th covers a clip two or four times as long within the 300-frame cap, at the cost of gaps in fast lights.
4. Pick the look. Light trails for anything that glows and moves; Smooth for water, clouds and crowds; Clear to remove people and cars that keep moving through a scene. All three are gathered in the same pass, so switching does not read the video again (Clear takes a moment to work out on a big picture).
5. Watch it build, then Save PNG (lossless) or Save JPEG (smaller). Cancel stops early and keeps the picture so far.

What the three looks do

Light trails keeps, for every pixel and colour channel, the brightest value any frame had there. A headlight that crossed the frame in 2 seconds leaves an unbroken line; the still, lit parts stay as they were. Smooth takes the average of all the frames for every pixel and channel. It is averaged in linear light, the way a camera sensor gathers it, and then turned back into ordinary sRGB: so a white light that sat on a pixel for half the frames comes out as 188 of 255, not the 128 a plain average of the numbers would give. Anything that stays still keeps its value; anything that moves fades by how long it stayed. Over 40 frames, a person who stood in one spot for 2 of them adds only 5 % of the light there, a faint ghost.

A worked example, the test clip: 2 seconds at 10 frames a second, black, with a white 10-pixel square crossing it, 10 pixels a frame. Light trails draws a solid white bar 200 pixels long. Smooth draws the same bar in dark grey, about 63 of 255, because each point along it was white in 1 of the 20 frames.

Clear (remove moving people) takes, for every pixel and colour channel, the median of the frames: the middle value once they are put in order (with an even number of frames, the average of the two middle ones). Whatever is at a spot for more than half the frames wins, so a tourist who walks across the square is outvoted by the paving behind them and vanishes completely, with no ghost. That is the difference from Smooth, which keeps a faint trace of everything in proportion to how long it stayed. A second test clip shows it: a grey gradient with a black 30-pixel square crossing it in the first 2 of its 5 frames (40 %). At 20 points along the square's path Clear gives back the gradient to within 1 of 255; Smooth leaves a darker band, 9 to 24 values deep. Clear keeps at most 31 frames; from a longer clip it picks 31 spread evenly from the first frame to the last, and the page says so.

Wikipedia's article on long-exposure photography describes the same effects: with a long shutter "vehicle lights draw bright streaks", "stars leave trails in the sky" and "water waves appear smooth", and it can "make moving people or dark objects disappear"; it depends on the camera being "stationary for the entire period of time that the shutter is open". Source: en.wikipedia.org/wiki/Long-exposure_photography, fetched 2026-10-29.

What it cannot do

• No alignment. Frames are stacked exactly where they are, so a handheld video blurs everything, not just what moved. Rest the phone on something or use a tripod.
• It opens only what this browser decodes. MP4 (H.264) and WebM open in every current browser. HEVC (H.265), the iPhone default, opens in Safari and only on some other browsers and devices; re-measured 2026-10-29, the headless Chrome this page is tested in reports it cannot play HEVC at all. When a file will not open, the page names it; convert it to MP4 (H.264) and try again.
• At most 300 frames, at most 1920 px on the long side. Measured 2026-10-29 in headless Chrome on a 4-core server: 300 frames of a 1080p 30 fps H.264 video took 80 s to stack (about 0.27 s a frame, almost all of it seeking and decoding), and a 320 × 240 test clip took 0.3 s for 20 frames. A phone is usually slower. For Smooth and Light trails memory does not grow with the number of frames: at 1920 × 1080 the page holds about 50 MB (a running sum, a running maximum and two picture buffers) whether it stacks 10 frames or 300. Clear also keeps up to 31 frames, at most 86 MB more (see below). So the cap is about waiting time: 300 frames is 10 s of 30 fps video at every frame, 20 s at every 2nd and 40 s at every 4th. Bigger video (4K) is stacked at 1920 px on its long side, and so is the saved picture.
• Clear removes only what keeps moving. It cannot remove a person who stands still, or anything that is on the same spot in more than half the frames: a queue, a parked car, someone who sits down for most of the clip. It cannot remove anything from one photo; it needs several frames from a camera that did not move. It is not AI: nothing is painted in, every colour value of every pixel is one a frame really had there (or, with an even number of frames, the average of two of them). At most 31 frames, at most 1280 × 720 px (921,600 pixels, so the kept frames stay under 86 MB); a longer clip gives 31 picked evenly, and a 1080p video is cleared at 1280 × 720. Measured 2026-10-29 in headless Chrome slowed 4× to a mid-range phone's speed: working out Clear from 31 frames at 1280 × 720 took about 2 s, and a 4-second 1080p clip (120 frames) took 44 s to read with Clear chosen against 40 s with Smooth.
• Light trails also brightens noise. The brightest value of every pixel includes the brightest speck of sensor noise, so a dark night sky turns grainier and lighter the more frames you stack. Smooth does the opposite and quietens noise.
• No RAW files. Photos must be ones the browser opens itself (JPEG, PNG, WebP; HEIC only in Safari). DNG and other RAW files will not open.
• Photos of other sizes are fitted to the first. Each later photo is scaled to fill the first photo's frame, centred, and whatever sticks out is cropped. Shoot the set at one size.
• Frames are read by time. Frame k is the one showing at start + k × the measured frame length (× 2 or × 4). Phone video often has a variable frame rate; a frame the browser shows twice is counted once.

Questions

How do I make a long exposure from a video?

Open the video on this page, drag the start and end handles around the part you want, and choose Use every frame (or every 2nd or 4th for a longer clip). Pick Light trails for car lights, stars, fireworks or sparklers, or Smooth for water, clouds and crowds. The picture builds as the frames are read; then Save PNG or Save JPEG. The video must have been filmed with the phone held still, because the frames are stacked exactly where they are.

Why is my whole long exposure blurry?

The camera moved while filming. This page does not line the frames up: it stacks every frame where it is, so if the phone moved 20 pixels, every edge is smeared over 20 pixels, the still parts as much as the moving ones. Only the things that moved in front of a still camera should blur. Film with the phone resting on something or on a tripod, or choose a shorter clip from a moment when it was still.

What is the difference between Smooth and Light trails?

Smooth averages every frame, so anything that moves fades in proportion to how long it stayed: running water turns to silk and a person who walked through disappears. Light trails keeps, for each pixel, the brightest value any frame had there, so every light that moved leaves a full-strength line: car lights, stars, fireworks. A bright light in one frame of 100 is a faint smudge in Smooth and a bright line in Light trails.

Is my video uploaded?

No. The browser's own player decodes the video on your device, each frame is read into a canvas on this page and added to a running sum and a running maximum, up to 31 frames are also kept in the page's memory for Clear, and the picture is written in the page. The only network traffic after loading is an anonymous page count and three fixed event names, longexp-open, longexp-save and longexp-clear-save.

Can it remove people from a photo?

Not from one photo. Clear needs several frames from a camera that did not move: a video shot on a tripod, or a burst of photos taken from the same spot. For every pixel it keeps the middle value of all the frames, so a person who walks through and is on that spot in fewer than half the frames is outvoted by the background behind them. From a single photo there is nothing behind the person to show, and this page does not invent it: it is not AI and paints nothing in. Someone who stands still for most of the clip stays in the picture.