Tips for the best results
A swap is only as good as what it has to work with. Two things matter most: the source video, photo, or GIF you're swapping into, and the replacement photo you provide for each face. Here's what actually moves the needle on both.
Your source media: lighting
Even, front-facing light is the easiest case — think daylight through a window, or any evenly lit room. Hard shadows across half a face (strong side lighting, direct sun) make it harder to reconstruct that side convincingly. Backlighting — a bright window or sky behind the subject, with their face in shadow — is one of the toughest cases, since there's less actual detail in the face to begin with.
Your source media: angle
A face looking roughly toward the camera — even a three-quarter turn — gives the swap much more to work with than a steep profile. It doesn't need to be a straight-on mugshot; natural angles from everyday photos and videos work fine. The cases to watch for are extreme angles (near pure profile), faces tilted far up or down, and faces during fast motion where blur sets in.
Your source media: resolution
A face that only takes up a handful of pixels in the frame — someone far in the background of a wide shot — simply doesn't have enough detail for a confident swap, and very small faces may not even be detected at all. If you have a choice between a compressed, heavily downscaled version of a video or photo and the original, the original almost always swaps better.
Your replacement face photo
• Front-facing or close to it — avoid a steep side profile for the replacement photo specifically, even if the source media has more natural angles.
• Even lighting on the face, without heavy shadows or a very bright background behind them.
• One face clearly visible and not obstructed by sunglasses, a hand, hair across the eyes, or another person's head.
• A neutral or mild expression tends to generalize better across a video's range of expressions than an extreme one (a wide-open laugh, for instance).
• Higher resolution is better — a small, heavily compressed thumbnail gives the swap less to work with than a clear photo.
For video specifically
Longer videos and videos with more people being replaced take proportionally longer to process — that's expected, not a sign of a problem. Scene cuts, very fast camera motion, and rapid zooming are the situations most likely to produce an occasional rough frame; a steadier shot tends to swap more consistently throughout.
For GIFs specifically
GIFs are typically short and low-resolution by nature, which is usually fine — faces in close-up reaction GIFs tend to be large in frame even at a small overall size. The same lighting and angle guidance above still applies; a clearer source GIF (less aggressive compression, a face that isn't tiny in a wide shot) gives a noticeably cleaner result.
If a result doesn't look right
Try a different replacement photo first — a more front-facing, evenly lit one is the single biggest lever you have. If the source media itself has the face partially hidden, badly lit, or very small, a higher-quality version of that source (if one exists) is the next thing worth trying.
← Back to Recast