The workflow
Choose a segment where the target stays identifiable
Watch the full clip before uploading. Mark where people cross, turn fully away, or cover one another. Start with a short continuous segment in which the intended person is visible. A target that repeatedly disappears behind someone else is a difficult test even with a precise prompt.
Use two stable anchors for the target
Combine a starting position with a persistent feature: “the person wearing the blue jacket, nearest the window at the beginning.” Position alone can become ambiguous after people move. Mention that the target is the same person throughout, and identify the other people as unchanged.
Add the replacement face and narrow the edit
Open Video Face Swap and add the reference as @Image1. Keep “Swap a face” selected. Describe facial identity separately from hair, clothes, body, and environment. The tool accepts MP4 or MOV under 64 MB and reference images under 10 MB. Use media you have permission to edit.
Review every person, not only the target
Compare the source and output during entry, exit, turns, crossings, and fast motion. Check the target’s identity, then look at all bystanders. Also check lip motion, hairline, jewelry, and background details. A correct opening frame does not establish that the swap remains correct later.
Recorded Made demonstration
What the recorded face-edit example demonstrates
This is a real single-person face edit from September 19, 2026: the reference supplies a new facial identity while the prompt asks to keep the copper bob, yellow shirt, and serum shot. It establishes the face-edit workflow, not multi-person tracking accuracy. The group-specific prompt below is an adapted recipe; no group success rate is claimed.
Kling O3 Edit Pro · Generated · View the example record and price basis (JSON)
Prompts you can inspect and adapt
Prompt recipe to adapt
Specify one person in a group
Replace only the facial identity of the woman wearing the yellow shirt, on the left at the start of @Video1, with the face in @Image1. Follow that same woman throughout the clip even when her screen position changes. Preserve her original hair, body, clothing, head motion, expression timing, and lighting. Keep the faces, clothing, bodies, and actions of every other person unchanged. Preserve the camera movement and background. Do not transfer the reference face to anyone else.Replace the description with details from your actual clip. Written targeting is not a tracking guarantee.
Recorded generation prompt
The exact single-person demonstration prompt
Replace only the facial identity of the main woman in @Video1 with the adult woman's face from @Image1. Match the reference face's facial proportions, features, skin tone and texture. Keep the original copper-bob hairstyle, yellow shirt, body, hands, emerald bottle, bathroom, lighting and background from @Video1. Preserve every head movement, expression, gesture, camera position and shot timing exactly. Only her face changes. Do not replace hair, clothing, product or environment; no beautification, cuts or text.The reference contains an outfit and setting too. The prompt explicitly limits the requested change to the face.
A five-second edit: 84 credits at the observed rate
5 seconds × 16.8 credits/second = 84 credits
On October 14, 2026, the public catalog returned a customer rate of $0.168 per input second for this model. Made displays one credit per cent. This is a dated estimate for one five-second clip at the tool’s default settings with original audio off; it is not the historical bill for the example.
Price source: Public video-editor catalog · Observed
The tool estimates using input duration rounded up to a whole second, then rounds the credit total up. A 5.1-second input at the same rate is estimated as 6 × 16.8 = 100.8, rounded up to 101 credits. Upload your actual clip to see its estimate; the backend probes the clip and confirms the price before generation. Each new generation needs its own quote.
What to review before you use the result
| What you see | What to try |
|---|---|
| A bystander receives the new face | Add a stable clothing or role description and shorten the segment before the people cross. Regenerate and review everyone; do not assume another sentence guarantees selection. |
| The target changes when people move | Describe who the person is at the start, then state “follow that same person throughout.” For a complicated crossing, edit shorter shots separately. |
| The reference hairstyle replaces the source hair | State “facial identity only” and explicitly preserve the source hairstyle. Use a clear face photo with a similar head angle when possible. |
| The face looks pasted on during a turn | Use a reference with visible facial structure and choose a less extreme turn for a first attempt. Check skin tone, lighting direction, face edges, and expression timing in motion. |
Work with Claude and Made
Claude can help write an unambiguous target description and prepare a Made video-editing request. Connect Made, let the assistant inspect the available editing model and media schema, and require a quote for the actual clip before generation.
Conversation starter
Ask your assistant
Use Made to edit only the face of the person in the yellow shirt, on the left at the start of my video. Use my attached face reference and keep everyone else unchanged. Inspect the available video-editing model and its input schema. Propose the target description, preserve the source audio, and show the exact credits before generating one result. If the target becomes ambiguous, tell me which shorter segment to use.The connector uses the same generative model limits. It does not add a manual face tracker or guarantee that other people will be untouched.
Questions about this workflow
Is there a clickable face selector?
This workflow selects a person through written instructions. It does not expose a click-to-lock face tracker or mask. Use a stable description and review the output for changes to other people.
Can I keep the original voice?
Keep original audio is available and enabled by default in Video Face Swap. A face edit does not require a new voice; review the downloaded clip to confirm the audio and timing you want.
What if two people wear the same clothes?
Add their initial location and role in the action, or choose a shot with a clearer target. If they cross or become indistinguishable, divide the edit into shorter segments instead of relying on position alone.