Add Text Watermark To Video.

Type the words, choose a face and a colour, place it, and export. The part that decides whether it works is contrast: plain white text disappears over a bright sky and plain black text disappears over a shadow, which is why the type here is drawn bold and stroked with a thin outline in the opposite tone before it is filled. Keep it short, because a watermark is a signature rather than a caption and sixty characters is already generous. Then set the opacity somewhere around seventy percent, which reads clearly as a mark without fighting the footage, and tile it instead of cornering it if the clip is one you expect somebody to crop and repost.

Yes. A PNG or WebP with an alpha channel is redrawn here as a PNG with that channel intact, and it is composited onto every frame rather than pasted over it, so only the pixels you actually drew land on the picture. That is why a round badge stays round and a wordmark with no background stays a wordmark. An SVG is rasterised first, because the video encoder deals in pixels, and it is rasterised large so it stays crisp even at sixty percent of a 4K frame. A JPEG has no transparency to keep, by design of the format, so a logo saved as a JPEG will arrive as a rectangle with whatever colour was behind it; if that is what you have, re-export it as a PNG first.

Always the same fraction of the picture. The size control is a percentage of the frame width, defaulting to eighteen, so on a 1920 wide clip that is 346 pixels and on a 1080 wide portrait recording it is 194. The margin works the same way, as a percentage of the shorter edge, so the mark sits the same distance in from the corner on a wide master and a tall reel. If a tall logo at your chosen size would run past the top and bottom margins, it is scaled down to fit rather than cropped, and the plan line says so. That is the whole reason for the tool: a fixed pixel size is the thing that makes a batch look wrong.

Up to forty in one go, which is the point rather than a bonus. They run one after another rather than all together, because forty simultaneous video encodes in one browser tab is a frozen tab; a queue with an honest bar is faster in practice and far easier to trust. The bar reads clip three of twelve and names the stage, each row shows its own status and its own size when it is done, and cancelling stops the queue where it is and keeps everything already finished. At the end the whole set is packed into one zip, stored rather than compressed, because an MP4 does not shrink and there is no sense spending seconds proving it.

Slightly, and unavoidably, because drawing on a picture means decoding it and encoding it again. The encode is aimed generously, at about twenty percent above the bitrate the compressor calls comfortable for that frame size and frame rate, and it is capped at the source's own bitrate so a marked clip is never larger than the clip it came from. In practice one generation at that setting is invisible on anything but a synthetic test pattern. The sound is not touched at all: the audio packets are copied across byte for byte whenever the container will carry them, so only the picture goes through the encoder.