Skip to main content

Edit character

Edit character is the Edit Video tool that keeps a video's original motion and timing but swaps in a new character. You give it a source video and a reference image of the character you want, and it re-renders the video with that…

L
Written by LX

Edit character is the Edit Video tool that keeps a video's original motion and timing but swaps in a new character. You give it a source video and a reference image of the character you want, and it re-renders the video with that character performing the same movements.

What does Edit character do?

It takes the motion from your source video and applies it to a different character. The movement, pacing, and roughly the framing stay the same — only the character changes to match the reference image (or your model) you provide. This is the mirror image of Edit motion, which keeps your character and borrows motion from elsewhere.

How do I run an Edit character edit?

  1. Open Edit Video and pick Edit character from the tool switcher.

  2. Add the video you want to edit with Select content or Upload video.

  3. Add the character you want using the reference-image controls (see below).

  4. Choose your quality and resolution in the bottom bar if you want to change them.

  5. Click Generate. The cost is shown next to the button.

How do I choose the new character?

You have three ways to supply the character:

  • Select content — pick an existing image from your Generation media (followed by curated inspiration) or Community media.

  • Generate reference image — open a dialog that creates a brand-new reference image for you (you can describe it, and it can use your source video as a guide).

  • Upload image — upload an image file from your device.

Once an image is set, it shows in the panel with a Delete button so you can swap it out.

What if I don't add a reference image?

With a portrait model selected, the Reference image remains optional: the panel notes, "If you don't provide a reference image, we'll choose one automatically," and skipping it uses that model. In No Model mode, the Reference image is required because there is no portrait model image to use. Generate stays disabled until you select, generate, or finish uploading one. Trying to continue without it shows "Please select or upload a reference image." An unfinished upload shows "The image is still uploading. Please wait a moment."

What is "Generate reference image"?

It's a built-in helper for creating the character image without leaving the tool. You type a short description, and it can use the first frame of your source video as a starting point so the new character fits the scene. When the image is ready you confirm it, and it drops straight into the reference slot. Generating a reference image is itself a small image generation, so it uses credits — the cost is shown when you generate it and depends on your plan and the quality you choose.

Is there a length limit?

Yes. Ultra and UltraS accept a selected clip between 4 and 30 seconds; Best accepts 4–15 seconds and supports 480P, 720P, and FHD. The hint below the source-video controls updates to "Use a video between 4 and 15 s." on Best or "Use a video between 4 and 30 s." on Ultra/UltraS. A shorter clip shows "The video must be at least 4 s." If a clip longer than 15 seconds is selected on Ultra or UltraS, switching to Best is blocked with "Trim the video to 15 seconds or less before switching quality." Use the scissors button next to the duration to shorten the selected slice, then switch again. Best is powered by Seedance 2.0, while Ultra and UltraS are powered by Seedance 2.5; hover a quality on desktop, or tap its info icon on mobile, to see the model note.

Can I jump straight to Edit motion?

Yes. When you're on Edit character, the panel shows a shortcut: "To make your character follow the video's motion, use Edit motion." Tapping it switches tools without losing your video.

On mobile

The flow is the same — the video and the character-image controls stack vertically, and Generate reference image opens as a full-screen page instead of a dialog. Pickers and uploads use full-screen sheets with progress and delete controls. In No Model mode the label is Reference image without "(optional)," and the automatic-image fallback note is hidden.

Did this answer your question?