# Video Recast

Part of the SocialArt.ai documentation. Index: https://socialart.ai/llms.txt · Open Video Recast: https://socialart.ai/tools/video-recast

Video Recast replaces the people in a video with the people from your reference photos. Pick the model that fits the job: Hailuo 3 Recast keeps the original footage and sound exactly as they are and only swaps the people; Seedance 2.5 and Wan 3.0 Prime re-render the scene from your video, at the cost of small differences in the footage and new sound: Seedance gives the highest overall quality, Wan keeps the faces closest to your photos. Open it from the Tools page, or pick Hailuo 3 Recast as a model in the Video Studio.

## Models

| Model | What stays the same | Source length | Resolutions |
|---|---|---|---|
| Hailuo 3 Recast | Footage, camera, cuts and the original sound | 5–30 s | 768p, 1080p |
| Seedance 2.5 | Motion, camera and cuts, re-rendered | 4–30 s | 480p, 720p |
| Wan 3.0 Prime | Motion, camera and cuts, re-rendered | 2–15 s | 480p, 720p, 1080p |

Hailuo needs every single shot to be 15 seconds or shorter. The maximum file size is 100 MB for Hailuo and Wan and 50 MB for Seedance. The generate button shows the cost of the chosen model and resolution before you start; Seedance and Wan bill the length of your video as input and as output.

## In three steps

1. **Source video.** Upload an MP4 or MOV, or pick one from your video history. The length and size limits of the chosen model are shown on the page (see Models).
2. **New people.** Add one photo per person that should appear in the video: upload, pick from your image history or choose a saved character (not available with Seedance). Up to 4 photos. By default the photos replace the main people in the video from left to right, in every shot they appear in.
3. **Generate.** Choose the model and the output resolution, optionally describe who becomes whom or what to keep or change, and hit Recast. The result appears on the page after a few minutes and is saved to your video history under the chosen model.

![The Video Recast page: the three steps on the left and the result on the right](https://socialart.ai/img/help/tools/video-recast-light.webp)

## Good to know

- **Cost:** billed per second of the source video and shown on the generate button before you start; a higher resolution costs more per second. Reference photos are free. Failed generations are refunded automatically.
- **Extra instructions are optional.** Use them to assign photos to people ("the man on the left becomes photo 2") or to keep details such as clothing or a hairstyle. With Seedance and Wan the tool always tells the model to replace the people in the video with the people from your photos, from left to right, and your text is added to that. Hailuo is built for this job and receives your text as you wrote it.
- **Best results** come from sharp, evenly lit photos that show the face clearly, and a source video where the people are fully visible.
- Hailuo keeps the source audio. Seedance and Wan generate their own sound for the re-rendered video.
- **Bypass audio restrictions (Seedance only).** Seedance sometimes refuses a video because of its sound. Switch the bypass on and the sound is removed before Seedance sees the video; the original sound is put back under the result for you. Leave it off unless a generation failed for that reason.

![The Bypass audio restrictions switch, shown when Seedance 2.5 is the chosen model](https://socialart.ai/img/help/tools/video-recast-bypass-light.webp)

- Results land in your video history, where you can download them, add them to albums, reuse them in the Video Editor or post them with Social Post.

## Related

- Saving characters: https://socialart.ai/llms/characters.txt
- Replacing a person in a photo: https://socialart.ai/llms/character-swap.txt
- All tools: https://socialart.ai/llms/tools.txt