Published · Sources checked October 27, 2026
If you only want a different face in an existing rap clip, start with face swapping. If the outfit, body, or character design also needs to change, look for character replacement. If you want a new character to follow a recorded performance, check performance-transfer controls. Each route gives the reference video a different job.
This guide explains documented controls and their limits; it does not rank the quality of generated results. If you have a portrait and a song but no reference performance, start with the photo-to-rap-video guide.
1. Choose what changes and what must stay
Write down the edit before choosing a tool: “change the face,” “change the character,” or “use this performance to animate another character.” The table describes the intended scope, not a promise that every frame will match your reference.
| Workflow | Intended change | Starting inputs | Check before choosing |
|---|---|---|---|
| Face swap | Facial identity in an existing shot. | Source video plus a replacement face image. | Face selection and consistency. Do not expect this route to redesign the outfit or body. |
| Character replacement | The subject’s wider appearance, such as character design or clothing. | Source video plus a character image in a replacement workflow. | Target-person controls, replacement scope, background changes, and continuity. Exact clothing or body detail is not guaranteed. |
| Performance transfer | A character’s performance, driven by another recording. | Driving performance plus a character image or video. | Which input supplies gestures, facial motion, and the scene. Image and video inputs can expose different controls. |
These categories can overlap in product menus. Magic Hour’s character-replacement guide distinguishes a facial-identity change from a wider character change and warns that a reference does not ensure exact appearance throughout a clip. Look at the selected mode’s controls, not just a “swap” label.
2. A motion reference is not always the scene you keep
A reference can be footage to edit, or a recording used to drive motion. Those are different roles. A performance filmed in your room does not mean the generated character will appear in that room.
In Runway’s Act-Two documentation, the driving performance and character input are separate. A character image supports gesture control and receives generated environmental motion. A character video instead supplies its existing subject, environment, and camera movement; Act-Two applies facial motion and expressions, with gesture control unavailable.
For this workflow, identify which file should define the final scene before uploading. If keeping a particular background or camera move matters, check the character input and inspect the output against it. “Follows the performance” alone does not establish what happens to the scene.
3. Match the workflow to your rap clip
The following are hypothetical editing briefs, not demonstrations of generated results.
Scenario A: keep the outfit and shot, change the face
You have a single-performer clip and want a different facial identity while keeping the existing visual setup. Start with the AKOOL Face Swap workflow. Its official instructions describe uploading the source, selecting a detected face, and providing a replacement face. Inspect the moving result rather than judging only the selection thumbnail.
Scenario B: change the character’s wider appearance
Your brief calls for a different character design, so a face-only edit would leave too much unchanged. Explore the Viggle workflow guide. Viggle describes Mix as combining a character image with a motion video and Multi as using several character images. Its separate Motion Control workflow animates a character in a static image. Choose the mode that matches your intended edit, then check the result’s replacement scope. These distinctions come from Viggle’s animation-tools description, not a guarantee of unchanged clothing detail or audio.
Scenario C: record a performance for a new character
You want your expressions and delivery to drive a character image or video. The Runway Act-Two guide covers that starting point. Choose the character input with the scene and gesture controls in mind. This is a performance-transfer brief; it does not require treating the recording’s original room as the final set.
4. Check people, mouth timing, and audio separately
People: seeing two performers in an input is not the same as independently assigning each person. Magic Hour says manual person selection depends on the model and mode, and trimming can invalidate selection points. Check that the intended person stays selected after any trim. AKOOL documents selecting a detected face, but that alone does not establish independent body or performance control.
Runway’s multi-character tutorial starts from single-character inputs, generates each performance separately, and composites the outputs in an editor. It is not a one-pass promise for every duo shot. For planning those separate parts, use the two-person rap-video guide.
Mouth timing: review the mouth against the intended vocal, including fast phrases, pauses, and hand-offs. A transferred body movement does not demonstrate that every syllable aligns. In a duo, check whether the silent person’s mouth moves during the other verse.
Audio: listen to the exported file. Identify whether it contains the source clip’s audio, the driving performance, a separately supplied song, or no track. Check that the intended vocal starts at the right point and that adding your master track does not leave duplicate audio underneath. Keep these checks separate from judging visual motion.
5. Review the whole clip before calling it finished
- Target: was the intended person changed, including after crossings or cuts?
- Identity: does the character remain consistent through turns and obstructions?
- Body and outfit: do they match the intended scope of the edit?
- Scene: are the background and camera movement acceptable for your brief?
- Mouth timing: do the visible performances follow the correct vocal parts?
- Sound: is the intended track present, aligned, and free of duplicate playback?
- Export: check the actual file, watermark, resolution, download access, and applicable usage terms before publishing.
Mark a mismatch by timestamp and by the control it concerns. That gives you a specific reason to change a setting, input, or workflow; it does not promise that another generation will fix it.