How to Face Swap a Video Online: Step-by-Step Guide (2026)

Video face swap workflow

Test a representative short segment first, then review the complete motion before processing more footage.

Video face swap uses a source portrait to replace a detected face while attempting to retain the target video's motion, lighting, and expression. Framedance accepts up to 8 videos per batch and supports single- or multi-face mapping. Occlusion, profiles, fast motion, and soft frames can still produce drift, so a completed status means processing finished, not that the clip is ready to publish.

TL;DR

Batch up to 8 videos in one run and swap faces across all of them at once.

Two modes: single-face (one source face per video) or multi-face (map every detected face individually).

Cross-video deduplication means the same person only needs to be mapped once, even across multiple clips.

Failed clips can be retried individually, and finished videos land in a history strip right on the page.

Review the displayed credit estimate before submitting; cost depends on the model, clip count, duration, and selected time range.

How do you swap faces in a video online?

1. Open Framedance's video face swap tool and sign in.
2. Upload your target videos. You can drag in up to 8 clips at once for batch processing.
3. Pick a mode: single-face if each video only needs one face replaced, or multi-face to map several faces individually.
4. Upload the source face photo(s). A sharp, front-facing photo with even lighting gives the best results.
5. If the same person appears in several videos, map them once — cross-video deduplication applies that mapping everywhere.
6. Optionally adjust the detection frame count for footage with fast motion or many people in frame.
7. Start the swap and watch per-clip progress. Finished videos appear in the history strip on the same page.
8. Preview and download your results, and retry any individual clip that failed — no need to rerun the whole batch.

Single-face vs multi-face mode: which should you use?

Single-face mode is the direct path: provide one source face per video and map it to the relevant detected person. It fits talking-head clips and other shots where only one identity needs replacement. Use multi-face mode whenever several visible people require explicit, separate mappings.

Multi-face mode detects every face in your footage and lets you map each one to a different source photo. Use it for group scenes, dialogues, or music videos. Each detected face gets its own mapping slot, so you always know exactly who becomes whom — and thanks to deduplication, a person who shows up in clip one and clip five still only needs a single mapping.

What can you use video face swap for?

Creators use video face swap for meme edits and reaction content, filmmakers use it to previsualize casting choices or fix a scene without a reshoot, and marketers localize one ad with different presenters for different markets. Batch mode is what makes the last two practical — you upload all the variants at once instead of babysitting clips one by one.

Framedance gives professional creators a flexible multi-model workflow. Whatever your use case, only swap faces you are authorized to use, confirm everyone is at least 18, and never use the feature for fraud or deception.

How much does it cost to face swap a video?

Video face swap uses credits, with the actual charge determined by the current model, clip duration, and processing range. Review the task cost shown before submission, validate mapping and quality on one short clip, and only then release the full batch.

Cost scales with the footage you process: each clip in a batch runs as its own job, so a short meme costs less than a long scene, and a failed clip can be retried on its own instead of paying for the entire batch again.

For automated batches, read the live API model detail and its current parameter and cost information; do not hard-code web-form fields or an old price into a client.

How do you get the most realistic result?

The source photo matters more than anything else. Use a sharp, front-facing shot with even lighting, no sunglasses, and only one face in the frame. A good source photo fixes more problems than any setting will.

On the video side, footage where the face is reasonably large in frame and not constantly motion-blurred swaps far better. For fast cuts or busy scenes, raising the detection frame count helps the tool track every face reliably.

Finally, iterate cheaply: run one clip first, check it in the history strip, and only then launch the full batch. If a single result disappoints, retry just that clip with a different source photo.

Can you automate video face swap with an API?

Yes. Create a key on /api-keys, read the current model parameters, submit input to that model's run endpoint, and poll the returned task_uuid until a terminal state.

Before writing the batch client, read the model's current parameter table and cost information, then validate the request and output shape with one small authorized task.

curl -X POST "https://www.framedance.ai/api/v1/marketplace/run/<model_id>" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "input": {
      "video_url": "https://example.com/clip.mp4",
      "face_image_url": "https://example.com/face.jpg"
    }
  }'

# Poll until the task completes
curl "https://www.framedance.ai/api/v1/marketplace/run/tasks/<task_uuid>" \
  -H "Authorization: Bearer YOUR_API_KEY"

Practice: prepare and map a difficult short clip

Choose an authorized clip that includes a front-facing moment, one moderate head turn, and a brief partial occlusion such as a hand passing near the face. Trim it to the shortest segment that still contains all three conditions. Prepare a sharp, evenly lit portrait of each authorized adult you intend to map; avoid sunglasses, heavy blur, extreme angles, and group photos as sources.

Use single-face mode only when one identity is being replaced. For a group scene, use multi-face mode and inspect representative detection frames before assigning each target identity to its source. Confirm every mapping, the selected processing options, and the displayed cost before submitting. Start with this one clip rather than a full batch so a poor source or incorrect mapping is inexpensive to discover.

Accept, diagnose, and retry

Watch the exported clip from beginning to end at normal speed, then inspect the turn and occlusion frame by frame. Accept it only when the intended person remains correctly mapped, facial edges do not visibly detach, skin tone and lighting remain plausible, and the identity does not jump at a cut. Review every mapped person; a convincing first frame does not validate the full video.

If identity is wrong throughout, correct the target-to-source mapping. If it fails only on turns, try a clearer source closer to that angle. If another person is captured after an overlap, choose better detection frames or simplify the test segment. If edges or lighting look poor everywhere, replace the source before changing other settings. Retry only the failed clip; extend the duration or start a batch only after the short test passes, because every submitted render may consume credits.