Video face retouch

Processing one frame takes hundreds of milliseconds, so a live preview would become a ragged slideshow pretending to be video. Hence two separate tools: a still frame with the wipe, to dial in the strength precisely, and “Check in motion”, a short stretch rendered into memory and looped. The play button on the timeline plays the source — it is there to find the right spot, not to judge the result.

The cost of one frame is measured on your device while the preview is prepared, then multiplied by the number of frames in the selected fragment. That is why the estimate appears before you click, not after. Order of magnitude: on a laptop one second of video costs roughly six to seven seconds of work, on a phone noticeably more.

A clip that long cannot be processed in a browser as a whole — that is hours. Take the piece you need with the handles on the timeline: only it is processed and saved, together with its audio. A fragment longer than ten minutes is refused up front rather than after the fact.

They stay untouched — retouching is applied only where a face is found. The share of recognized frames is written on the frame itself: if the person turns away or leaves the shot that is normal, but a small share usually means the face is too small for the geometry.

By default yes, each face gets its own mask. For interviews and call recordings there is a switch for “only the largest face”: a defocused second person or a portrait on the wall gets a mask built from blurry geometry, and that shows more than the retouch helps.

There is a physical limit. A highlight is the color of the light source on top of the color of skin, and the two can only be separated while they still differ. If the patch is blown to white, no data about the skin underneath is left in the file — there is nothing to restore, and saying so is more honest than painting in a gray blob.

No: the output frame rate is taken from the source rather than forced to a common number, and the audio track is carried over as is and cut exactly at the fragment boundaries. Earlier the output was always assembled at 25 frames per second, and a 30 or 60 fps recording lost its smoothness of motion.

The one picked in the list; by default the source resolution. The clip is never upscaled: offering 1080p for a 720p recording would be selling a stretch. Which options your plan allows is visible right in the list.

It raises or lowers light only inside the skin area, leaving background and clothes as they were, and it does so with a soft curve: black stays black, white stays white, the midtones move. Plain brightness would blow out exactly the forehead the shine was just taken off. Useful when a window or a lamp ended up behind you.

MP4 and WebM are both offered. MP4 with H.264 opens everywhere — in editors, on social networks and on phones; WebM is for a lighter file on a website. The file is assembled by the browser’s hardware encoder, which is noticeably faster than a software one.