How to Make a Deepfake Online: Step-by-Step Guide

1. Choose a format. Start with a photo for a quick result. If you need a clip, prepare a video with a stable frame.
2. Prepare the target. Upload a photo or video of the person whose face will be replaced.
3. Choose the donor face. It can be your own portrait or an image you have the right to use.
4. Run generation. The network aligns the faces, transfers features and restores details.
5. Review and download. Check expression, lighting and face edges. If needed, change the source and repeat.

• The face occupies a noticeable part of the frame and is clearly visible.
• No strong blur, noise or compression artefacts.
• Even lighting and natural skin tone.
• The head is not covered by hands, hair or a mask.
• In video, the person looks at the camera and moves moderately.

For the donor, use several portraits from different angles — this helps the model capture the features more accurately.

• Too dark or overexposed photo. The network cannot see details and adds artefacts.
• Strong head tilt. Facial features get distorted and the swap looks unnatural.
• Mismatched donor and target angles. The closer the head positions, the better the match.
• Expecting perfection from a weak source. Result quality is limited by input quality.

If the face looks unstable, try a sharper frontal portrait with a neutral expression.

Photo processing usually takes a few minutes; video takes longer because the network processes every frame. Time depends on clip length, resolution and server load.

To avoid wasting attempts, test the swap on a single photo first: you will see whether the sources work and only then move on to video.