How face swap works
"Face swap" sounds like one single trick, but it's actually a short pipeline of distinct steps running one after another. Understanding what each step does makes it much easier to predict when a swap will look great, when it won't, and what to try if your first attempt disappoints you.
1. Finding every face
The moment you upload a video, photo, or GIF, Recast scans it for faces — not just one, all of them. A birthday photo with six people gets six detected faces; a video where someone turns away from the camera and back gets tracked as the same person across that whole stretch, not treated as a new face each time they reappear. Each detected person is numbered (Person 1, Person 2, and so on) so you can tell them apart when it's time to choose replacements.
2. You choose who to replace
This is the part that's entirely up to you. Every detected face gets its own thumbnail, and you decide which ones — if any — get a replacement. You don't have to replace everyone: swap one face in a group photo and leave the rest exactly as they were. If you don't pick a replacement for someone, Recast never touches their face at all.
3. The replacement face
For each person you want to replace, you upload a clear photo of the new face. Recast reads the distinguishing structure of that face — not a literal cut-and-paste image, but a mathematical description of its shape and features — so the swap can reproduce that structure convincingly under the original photo or video's lighting, angle, and expression.
4. The actual swap
With a target face chosen, Recast regenerates that part of each frame — blending the replacement face's identity onto the original person's pose, expression, and lighting. For a video, this happens frame by frame, many times per second, which is why longer videos and videos with more replaced faces take longer to process than a single photo.
5. Watermarking
Every single result — no exceptions — is labeled as AI-modified before you get it back. If the original had a cloned voice too, that gets an inaudible audio watermark as well. The goal is that a Recast result is never mistaken for unaltered footage, by you or by anyone you share it with.
Why some swaps look better than others
• A face that's small, blurry, at a steep angle, or partly covered is harder to read accurately — the swap has less information to work with.
• A clear, front-facing, well-lit replacement photo gives the swap a much stronger reference than a dark, cropped, or angled one.
• Fast motion or extreme expressions (mid-blink, mid-laugh) in the source video are naturally harder to match convincingly, frame to frame.
Our Tips for the best results guide goes deeper on exactly what to look for in both your source media and your replacement photo.
Consent, every time
None of this works without permission. Before a swap starts, you confirm that everyone involved — the original person and the replacement — has agreed to it. That's not a formality; it's the line between a fun, consensual edit and something Recast exists specifically to prevent. See our Terms of Service for the full rules.
← Back to Recast