How to Add a Watermark to a Video (Free)

Open the [free video watermark tool](/watermark-video), drop in an MP4, MOV, WebM or MKV up to 2GB, type your text or upload a PNG logo, pick a size, an opacity and one of nine positions, then click Add watermark and download the MP4. Everything runs inside the browser tab through ffmpeg.wasm, so the file never crosses the network and there is no account to make. A preview draws your mark over your own clip before you commit, so you can tell whether it is too faint or too large without exporting first. Nothing from PixelTools is stamped on the result: the only mark on the video is the one you chose. That is worth checking elsewhere too, since the free trial of at least one popular desktop watermarking app adds its own line of text to every file it exports.

Very little, because both modes draw onto the same transparent layer and share every other setting. Text mode takes a line such as a name or a copyright notice and renders it bold, with a soft drop shadow, in white, black or grey. Logo mode takes a PNG, WebP or SVG and draws it at the scale you set. The detail that matters is that size is measured as a share of the video's width rather than in pixels: text runs from 1 to 20 percent, a logo from 5 to 50 percent. A mark set at 5 percent therefore looks the same on a 720p phone clip and on a 4K export, which is exactly what you want when marking a batch of clips shot on different cameras. A PNG with a transparent background still gives the cleanest result, for the same reason it does when you [add a watermark to an image](/blog/add-watermark-to-image).

Usually it makes it smaller, which catches most people out. Running the exact filter graph this tool uses over the repository's own 1920x1080 test clip turned a 10.04 MB file into a 6.86 MB one, about 31 percent smaller, with the duration unchanged at 47.317333 seconds and the frame count unchanged at 1419. The command was `ffmpeg -i in.mp4 -i overlay.png -filter_complex "[0:v][1:v]overlay=0:0:format=auto,scale=trunc(iw/2)*2:trunc(ih/2)*2,format=yuv420p[v]" -map "[v]" -map 0:a? -c:v libx264 -crf 23 -preset veryfast -c:a copy out.mp4`. The picture has to be re-encoded because every frame changes, and a second H.264 pass at CRF 23 over footage that was already compressed once has far less fine detail left to spend bits on. Your own result depends entirely on the source, and [bitrate](/blog/what-is-bitrate) is what decides it. If the output is still too large for where it is going, [compress it afterwards](/blog/compress-video-online) rather than before, so the mark goes onto the better copy.

For MP4 and MOV sources, nothing at all. The audio stream is copied straight across, and in the measurement above the MD5 of the audio packets was identical before and after, d96350edcd657e8b50f9eeea28e5d6f7. That is deliberate: a watermark has nothing to do with sound, so a needless re-encode would cost quality for no reason. WebM and MKV are the exception. Their audio is usually Opus or Vorbis, and ffmpeg will quite happily mux either into an MP4 container, but Safari, QuickTime and most phones then play the video in silence. So those two formats go straight to AAC at 192 kbps instead. There is one more safety net behind that rule. If the copy attempt exits non-zero anyway, the tool deletes the half written output, resets its progress tracker so the retry's duration lines are not counted twice, and runs the whole job again with AAC rather than handing you an error.

A clip shot sideways on a phone ends up with the mark in the corner you picked, not in a rotated one. Vertical video is normally stored as a landscape frame plus a rotation note in the metadata, which players apply on the way to the screen. Before drawing anything, the tool runs a probe pass, reads the stream table ffmpeg prints for itself, catches the `rotation of -90.00 degrees` line and swaps width and height, because ffmpeg has already turned the picture upright by the time the overlay filter sees it. Without that swap a 1080x1920 portrait clip would get an overlay built for a 1920x1080 frame and the mark would land nowhere near the corner. It is the same metadata that makes [videos turn sideways after converting](/blog/video-sideways-after-converting), handled rather than ignored.

Bottom right is the default, because that is where broadcasters and stock libraries put a bug and it is the least intrusive of the nine positions on the grid. The grid covers four corners, the middle of each edge and the centre. A corner mark stays out of the picture but is the easiest to crop away; a centre mark survives cropping but sits on top of whatever you shot. The margin is not a fixed pixel count. It is 3 percent of the frame's shorter side, so the gap from the edge looks the same on a square clip and on a widescreen one. Opacity runs from 10 to 100 in steps of 5 and starts at 70, heavier than the 30 to 50 that suits a still photo, because a moving picture hides a faint mark far more easily. The same corner versus centre judgement applies when you [add a watermark to a PDF](/blog/add-watermark-to-pdf).

Two things do not behave quite the way the interface implies. First, the preview overlay is rendered on a 640 pixel wide canvas and the text size has a hard floor of 8 pixels. At the slider's 1 percent minimum the preview draws 8 pixels where the proportion asks for 6.4, so the preview mark is roughly a quarter larger, relative to the frame, than the one you download. Every setting from 1.5 percent upward is exact, so nudge the slider one step up if the preview is the thing you are judging by. Second, the logo picker accepts SVG, but an SVG carrying only a viewBox and no width or height attribute has no intrinsic size for the browser to read. It passes the file picker, the thumbnail beside the picker even renders, and the job then fails at process time with a generic message. Export that logo as a PNG and [run it again](/watermark-video).