Upload a video and character photos to give a familiar performance a new cast. Use the original actions, timing and camera movement as your guide.
The motion preview is an AI-recast display copy. The 15-second result used the original clip and this photo.
An AI character swap video replaces an on-screen person or character using reference photos. Unlike a face swap, it can guide the character's hair, clothing and body appearance as well as facial features. The video supplies the performance and camera reference; the photos define the new cast. This is generative video editing, not timeline editing.
The result is a newly generated video, not a pixel-perfect modification of the original. Check the whole performance, especially turns, hands, overlapping people and background details.
Seen a familiar dance, rap duet or performance with a different cast on social media? Character swapping is one way to make those remakes. Templates preselect the scene and roles; this tool lets you bring your own clip.
Street-performance remakes focus on the featured performer among bystanders. Specify the main character so the person holding the speaker and the surrounding crowd are not replacement targets.
Format source: N TelemicroThe orange-studio duet gives two people distinct roles in one recognizable setup. Assign each photo to a performer rather than letting the model infer who replaces whom.
Format source: AP NewsOne person plays the keyboard while another directs and reacts. Their jobs differ even when both faces share a frame: map photos to the keyboard player and the standing performer.
Format source: RedditSources explain these formats, not which model made every social-media edit. Our How to panel separately identifies an actual generated example and its input photo.
Choose a short video with the actions you want to reuse. Start with a clearly visible performer and simple camera movement. Trim to the scene you want before uploading.
Upload a clear photo for each replacement character. For a duo or a selective swap, turn on Prompt and name who each image replaces. State which other people should stay.
Choose the length and resolution, then review the credit total. Watch the finished clip for identity, clothing and motion consistency before downloading it from your result or history.
Similar inputs can serve different goals. Choose by what you want to replace, not just by whether a tool accepts a photo and a video.
| Approach | What changes | What the video provides |
|---|---|---|
| Character swap | The subject's face, hair, clothing and body appearance, guided by photos and instructions. | The performance, timing, scene and camera reference for the new cast. |
| Face swap | Primarily facial identity; the original body and wardrobe usually remain. | Existing footage supplies the body movement and scene. |
| Motion transfer | A reference character performs borrowed movement; the scene may be rebuilt. | Movement guidance, not necessarily a replacement inside the original scene. |
Use the image numbers shown in the tool. Identify performers by position or role, and keep that assignment across shots. These are starting prompts, not separate generation modes.
Replace the lead performer in Video 1 with the person in Image 1, including their hair and outfit. Preserve the original actions, timing, camera movement, background and other people.
In Video 1, replace the keyboard player with Image 1 and the standing performer with Image 2. Use each photo's appearance and outfit. Keep their roles consistent and preserve the original actions, timing, camera movement and scene.
It can guide the whole character, including hair, clothing and body appearance. Show the intended outfit in the photo. If you want the source video's clothes instead, say so in Prompt. Clothing transfer is not guaranteed frame by frame.
You can supply separate photos and assign each to a named role. For a selective swap, explicitly name the people to leave unchanged. More reference images do not guarantee more stable characters; overlap and cuts make identity mapping harder.
Upload a 2–15-second MP4, MOV or WebM video and 1–10 JPG, PNG or WebP images. Choose 480p or 720p output, no longer than your reference clip. The upload controls show file-size and image-dimension limits.
No. Prompt is off by default and uses general recasting instructions. Turn it on for precise role mapping or clothing choices. Use Video 1 and the numbered images. Turning it off means your saved custom text is not sent.
Use Character Swap for your own video. Choose a Trending template when you want its preset scene and role mapping. Both use visual references, but a template's sample does not establish the result you will get with a different clip.
The total includes reference-video processing and output. A 15-second reference plus 15-second result costs 225 credits at 480p or 375 at 720p. Check the displayed total before submitting. Each new generation is a separate attempt.
They guide generation, but are not pixel-locked. Request the original actions, timing, camera and background, then check the result for drift. Fast turns, occlusion, hands and multi-person scenes can change. This is not a guarantee of frame-for-frame reconstruction.
The current workflow requests audio from the generation model; it does not copy the source soundtrack unchanged. Listen for changes to speech, music and timing. Do not assume the result includes the original recording or exact lip sync.
You can try clear pet or original-character references, but human motion may distort different anatomy. Start with a simple clip and check limbs, proportions and appearance throughout the result. Pet performance has not been validated for this custom workflow.
The same reference workflow can also guide an outfit, background or product change. Turn on Prompt, identify the target and assign its image. Inspect these edits separately: labels, small objects and hand contact can change unexpectedly.
Browse dance and performance templates with their reference clips, roles and examples already selected.
