Precise video editing

Precise editing changes a subject, background or audio in an existing clip. Identify one editing master and state exactly what changes and what remains.

Start an edit

1. With Seedance 2.5, choose Video editing (视频编辑) at the upper-right of a shot video. Canvas video nodes also offer an entry.
2. Select the source and trim the relevant segment if needed. For stability, prefer a source of 20 seconds or less; longer edits may need more iteration. This is quality guidance. Model reference-duration limits and the current upload validation still apply.
3. Capture a relevant frame or upload target subject, object or background references.
4. Replace the template placeholders in the prompt editor. Describe the scope and the change from A to B.
5. Generate a candidate and compare with the original, especially areas that should remain unchanged.

Open precise editing from a generated shot.

Open precise editing from a generated shot.

Select a segment, prepare references and specify the edit.

Select a segment, prepare references and specify the edit.

CoolShowAI

Subject replacement template

[Goal] Edit @Video1, replacing only the black coat with the beige trench coat in @Image1.
[Master] @Video1 defines the scene, camera, movement, action, occlusion and event order.
[Target] @Image1 defines only the coat's color, design and material.
[Preserve] Keep identity, expression, body movement, background and audio unchanged.
[Detail] The coat folds naturally with movement and matches the original light and occlusion.

Subject-edit output example: inspect movement, contact and occlusion around the changed subject.

Background replacement template

Edit @Video1, replacing only the view outside the window with the rainy street in @Image1.
The original defines people, foreground objects, composition, camera movement and event order.
Match the new background's perspective and depth of field to the original camera.
Integrate reflections and rainy-night lighting naturally.
Preserve appearance, clothing, actions, dialogue and clip duration.

Background-edit output example: compare perspective, lighting and foreground edges.

Edit audio

State whether the task removes music, replaces dialogue or changes language. “Remove music while retaining dialogue, ambience and visuals” is clearer than “improve the sound.” For another language, provide the exact text, speaker, accent and emotion; inspect lip sync and residual original audio.

Limit the scope

For a partial edit, identify the time interval and region. Prefer 1–5 relevant images rather than contradictory definitions of the same object. Separate complex tasks into stages: subject, background, then audio.

Acceptance checks

The correct object changed; unrelated content remained stable; no temporal jumps; plausible occlusion and contact; complete dialogue; synchronized audio. If the result fails, narrow the requested edit instead of adding conflicting requirements.

Preserve count, trajectory and timing

Beyond A → B, state the edited object's count, entrance and exit times, trajectory, speed, contact and occlusion. If multiple videos are referenced, identify exactly one editing master. Other clips should supply only their assigned appearance or sound, without reorganizing the story.

A replacement background should develop correct perspective and parallax as the camera moves. Avoid describing a moving-shot environment as a static flat plate. For audio-only edits, explicitly retain or replace each category: dialogue, ambience, action effects and music.