How to Replace Faces in a Video With a Consented Reference Photo

Unseen's face-swap workflow takes a video and one replacement photo, then replaces every detected face in the video with that face.

It is cloud processing, not an offline feature. It is also an all-faces operation: you cannot use this workflow to swap one selected guest while leaving other detected faces unchanged.

Start with a short clip of a consenting adult and a reference image you have permission to use for the resulting depiction.

Prepare the video and reference photo

The current swap limit is four minutes, with a cloud upload limit of 500 MB. A longer video needs to be trimmed in another editor first. For a first test, choose a short section containing a head turn and some speech rather than processing the whole production.

Use a clear, front-facing reference photo of one person. The photo chooser accepts JPEG, PNG, and WebP and limits the image to 12 MB. A group photo makes the intended reference less clear; choose a suitable single-person image instead.

Check the video for background people before uploading. They are part of the all-detected-faces operation too.

Choose Face swap and add the photo

Open Unseen, sign in for cloud processing, and choose Face swap as the method.

In the Replacement face section, select Upload a photo. Check the displayed reference. If you picked the wrong image, use Choose a different photo before starting the job.

The interface needs a replacement photo before it can start a swap. If the chooser rejects the file, read its error rather than repeatedly trying the same oversized image. Create a suitable smaller copy in your image editor if necessary.

Select the video and use Swap all faces. There is no face-selection step for this workflow.

Inspect the difficult moments first

After processing, review the beginning, head turns, hands crossing the face, and any cut to a different shot. Then watch the complete export.

Look for three separate problems:

• Exposure: a face was missed or the original appears briefly.
• Replacement quality: the result distorts or changes abruptly.
• Scope: a background face was replaced when it should not have been in the production.

A reference photo that works in one shot may not solve every pose or obstruction. If the result is unsuitable, try a better source shot, shorten the edit, or use ordinary masking instead. Unseen does not currently provide a manual frame-repair control in this workflow.

Export with an explanation

Use Export video and review the downloaded file, including its sound. Face replacement does not anonymize the speaker's voice or remove names in captions.

For a staged demo, a simple title or caption can explain that facial appearance was digitally replaced. Keep that explanation with short excerpts and thumbnails too.

Retain the reference image only as appropriate for the project and the permission you have. Do not infer a cloud deletion schedule from the fact that the finished video has downloaded.