Seedance 2.5 video editing and extension: a prompt guide

The most expensive habit in AI video is re-rolling a clip you already like. The lighting was right, the performance landed, the camera move finally behaved — and then you regenerate the whole thing because one prop is the wrong colour, or because the shot ends three seconds too early. Every re-roll is a fresh dice throw against everything that was already working.

Seedance 2.5 video editing and extension exist to break that habit. Both are live in the OmniArt video workspace: pick Seedance 2.5, hand it a video you already have, and change one thing inside it — or add time to either end of it — instead of starting over.

The discipline behind both is the same, and it is unusually explicit. Name one editing master. State the edit scope narrowly. Spell out what must not change. And for extensions, describe the boundary frame before you describe anything new. Get those four right and the model stops treating your prompt as an invitation to reinterpret the shot.

What gets locked before you type anything

Editing and extension both inherit settings from the video you upload, and this surprises people the first time. You cannot pick the aspect ratio, and for editing you cannot pick the duration either.

Two practical consequences. First, if you need a different aspect ratio, editing will not give it to you — that is a reframing job, and the video frame expansion workflow covers it separately. Second, an edited result that comes back a fraction of a second longer or shorter than your source is expected behaviour, usually from transition-frame handling, not a failed generation. Content and event order stay substantially the same.

Part 1: editing a video you already have

Editing covers everything that changes what is already inside the frame: swapping a subject, replacing a background, stripping a music bed, cleaning up a stray object. The patterns below share one spine — declare the source video as the only authority, then narrow the change until the model has nowhere left to improvise.

The general editing pattern

Every editing prompt worth writing has five blocks. The first two do the heavy lifting: they tell the model which material is authoritative and which is merely a swatch.

[Edit Goal]
Edit @Video 1. Within <the entire video or a specific time range>, <add, remove, replace, or adjust> <visual object, region, or audio category>.

[Source Video Role]
@Video 1 is the sole editing master. It defines <characters, scene, actions, composition, camera movement, occlusion relationships, audio, and event order>.

[Target Material Role]
@Image 1 or @Audio 1 defines <specified attributes of the target object or sound>.

[Edit Scope]
Modify only <object, region, time range, or audio category>.

[Content to Preserve]
Keep <visual content, motion, audio, and timing relationships that must not change> from @Video 1.

Worked example — a timed lighting change with no other consequences:

[Edit Goal]
Edit @Video 1. Only from 4-7 seconds, change the cool blue light on the right wall to warm orange light.

[Source Video Role]
@Video 1 is the sole editing master. It defines the character, room layout, actions, composition, camera movement, audio, and event order.

[Edit Scope]
Change only the light color on the right wall and the area it illuminates. Allow the character's skin tone to respond naturally to the environmental light.

[Content to Preserve]
Keep the character's identity, clothing, expression, position, motion, room structure, camera movement, dialogue, and ambience from @Video 1.

Note what the edit scope does there: it grants one deliberate spillover (skin tone reacting to the new light) and refuses everything else. Naming the permitted side effect is what stops the model from either ignoring it or using it as licence to relight the whole room.

Subject replacement and timeline inheritance

Swapping an object is where "sole editing master" earns its keep. A reference image of the replacement object carries a background, a composition and a light direction you did not ask for, and the model will happily import all of it unless told not to.

Subject replacement adds a fifth block, [Timeline Inheritance], which is the part most people skip and then regret.

[Edit Goal]
Edit @Video 1. Replace only the yellow folding desk lamp with the white folding desk lamp in @Image 1.

[Source Video Role]
@Video 1 is the sole editing master. It defines the desk, books, hand movements, camera position, camera movement, occlusion relationships, and event order.

[Target Reference Role]
@Image 1 defines only the white folding desk lamp's appearance, structure, and material. Do not use the image's background, composition, or other objects.

[Edit Scope]
Keep exactly one white folding desk lamp throughout the video. Replace only the original yellow folding desk lamp. Do not modify the books, desk, hands, or background.

[Timeline Inheritance]
The white folding desk lamp inherits every appearance, lamp-arm rotation, hand occlusion, and exit of the original yellow folding desk lamp, including timing, path, and speed changes.
Except for the object or area explicitly modified above, keep all other people, props, scene content, camera movements, cuts, and event order from @Video 1 unchanged.

Timeline inheritance is a single sentence that says: the new object shows up when the old one showed up, moves where it moved, gets occluded when it got occluded, and leaves when it left — at the same timing, on the same path, at the same speed. Without it, replacements tend to appear late, drift off the original motion path, or survive a moment the original exited.

Background replacement

Background replacement is subject replacement inverted: the silhouette is the boundary, and everything inside it is protected.

The rule that makes it work is a tight reading of the reference image's job. @Image 1 supplies spatial layout, materials, depth of field, ambient colour and lighting direction — and nothing else. Say so explicitly, including "do not use the people or foreground objects in the image."

[Edit Scope]
Modify only the background outside the subject's silhouette. Do not modify subject identity, facial features, hairstyle, clothing, expression, position, size, or motion.

[Timeline Inheritance]
Keep the character actions and occlusion relationships from @Video 1. Except for the object or area explicitly modified above, keep all other people, props, scene content, camera movements, cuts, and event order from @Video 1 unchanged.

Keeping occlusion relationships in the preserve list matters more than it looks. When a hand passes in front of a body or a chair crosses the subject, the new background has to respect the same depth order — otherwise the swap reads as a sticker.

Audio editing

Dialogue, language, voice, background music and sound effects are edited separately from each other and separately from the picture. The structure is unchanged: state the speaker or sound category, the intended change, and which other sounds stay put.

Edit @Video 1. Remove only the original background music. Keep the character dialogue, lip sync, ambience, and action sound effects; preserve the visuals, camera treatment, and editing rhythm from @Video 1.

Edit @Video 1. Change <Presenter>'s spoken language to natural American English while preserving the dialogue content and speaking times. Keep all other character voices, background music, ambience, and visuals from @Video 1.

BGM removal is the standout here. Seedance 2.5 can strip background music while keeping the speaking voice, and — the part that usually breaks in other tools — it preserves on-screen subtitles and other visual elements untouched while doing it. That makes "clean vocal track, original picture intact" a single generation rather than an audio-tool round trip.

Smart Edit, Edit with marks, and Edit Video

Three related surfaces, one formula between them:

Plain Smart Edit takes [Specific modification object/what] + [Action and change description] + [Effective time sequence]. Annotation editing prepends one more block:

[Annotation tool and orientation] + [Specific modification object/what] + [Action and change description] + [Effective time sequence]

The annotation tools are boxes, lines, arrows, text and coloured landmarks, plus an eraser. The block that people under-write is orientation: you must say which tool you drew with and whether the model should act inside or outside it. "Within the red box." "The area the red arrow points to." "To the left of the red line." "At the landmark point." Without that, a box is ambiguous — protect this, or change this?

Useful verbs to reach for: "Edit the xx in the video", add, delete/remove, modify/replace/change to. Keep timing explicit in the same breath, as in "modify everything entirely from 0 to 8 seconds."

Partial elimination and re-angling a shot

Two capabilities worth knowing about before you decide to regenerate:

Partial elimination and traceless erasure. Brush or box a region and remove what's in it. The model recompletes the background according to physical and lighting logic rather than smearing, so clutter in a corner, a stray passer-by, or a prop you no longer want is a local fix. It works in the other direction too — directional element replacement drops a new element into the same region with the original video's perspective and atmosphere already applied.

Space perspective modification. Because the model reads 3D spatial relationships in the source, you can re-angle a shot you already have: overhead, low angle, first-person, or "replace the multi-camera switching shots with a fixed frontal shot, and keep the characters and background unchanged." Proportions and layout stay coherent through the change. That turns a competent-but-flat take into a directed one without re-rolling the performance.

Part 2: extending a video you already have

Extension is the other half of not starting over. Any result of 30 seconds or less can be extended, and extensions stack — you can keep adding to the result, nesting-doll style, up to 60 seconds of continuous video. Aspect ratio stays locked to the source, but unlike editing, duration is yours to set.

The prompting rule is one sentence long: align the boundary frame before you describe anything new.

Forward extension

For a forward extension, the extension's first frame continues from the source's last frame. So describe that last frame's continuous state first, then what happens next.

@Video 1 is the source video to extend forward.

Extend @Video 1 forward. The first frame of the extended segment directly continues from the last frame of @Video 1. Maintain continuity in <subject pose and orientation>, <prop position>, <background and spatial relationships>, <camera position and composition>, <lighting>, and <motion direction>.

Then, <describe the new action, event, camera treatment, or audio to add>.

Throughout the extension, maintain continuity in <character identity and clothing>, <key props>, <background layout>, and <axis of action>.
Keep each subject as the same continuous instance throughout: do not duplicate or split it, and keep the person's appearance or the object's number of parts stable.

Filled in:

@Video 1 is the source video to extend forward.

Extend @Video 1 forward. The first frame of the extended segment directly continues from the last frame of @Video 1. Maintain the same locked-off medium shot, the orange paper airplane's position and orientation, the classroom-window background, the afternoon lighting, and its movement toward the right side of the frame.

Then, the orange paper airplane continues gliding toward the right and exits the frame while the white curtain beside the window sways slightly. Keep the camera and classroom background in the state established by the source video's last frame.

Backward extension

Backward extension builds the segment before your source. Describe what happens first, then define the source's first frame as the explicit end state of the new segment.

That second half is the part that fails when you skip it. Writing only "then connect to the source video" leaves the model two ways to go wrong: characters or effects that belong to the source enter the preceding segment too early, or the picture keeps changing after it has already reached the target state.

@Video 1 is the source video to extend backward.

Extend @Video 1 backward. Before the source video begins, show an empty establishing shot of the same glass greenhouse. Morning mist drifts slowly near the floor, the overhead shade rises gradually, and no people are present yet.

The last frame of the extended segment naturally connects to the first frame of @Video 1. Match the greenhouse's central aisle, planting tables on both sides, glass frame, soft morning light, and locked-off wide composition. At the end, the shade is fully raised, the aisle is empty, and the leaves still sway slightly.

Extensions with additional reference materials

If you bring images or audio into an extension, define every material's role first, then state that the source video controls the boundary. New materials may supplement characters, props or audio — they must not override the source's last-frame (or first-frame) control of the join.

For backward extensions specifically, add one more line naming which materials must appear only after the source video begins:

@Image 1 defines <Curator>'s facial features.
@Image 2 defines <Curator>'s dark blue work jacket.
@Image 3 defines <Wooden Display Case>'s structure and material.
@Video 1 is the source video to extend backward.

Extend @Video 1 backward. Before the source video begins, <Curator> walks to the workbench, picks up the closed <Wooden Display Case>, and opens its lid.

The last frame of the extended segment naturally connects to the first frame of @Video 1: <Curator>'s pose and orientation, the case's position and state, and the other characters' positions. Match the camera position and composition, lighting, and motion direction of @Video 1's first frame.

<Materials that should appear only after the source video begins> must not appear early in the backward extension.

Naming material roles is the same skill the whole model rewards; the Seedance 2.5 reference prompting guide goes deeper on that side.

Extension vocabulary and constraint phrases

The model responds to specific terms. Use "forward extension", "backward extension" and "continuation" rather than describing the intent in your own words. For a whole-shot extension that must feel like one continuous take, add the constraint phrases directly:

Require natural extension, smooth movement connection, prohibit rigid cutting of the shot, prohibit objects appearing out of thin air.

Extension also unlocks segmented direction — run the character for ten seconds, confirm it, then instruct the jump in the next extension. Complex actions that fail as one long generation often succeed as a relay of confirmed stages.

Getting started on OmniArt

Pick one clip you already like and one thing about it that is wrong. Open the OmniArt video workspace, select Seedance 2.5, upload the clip, and write the five blocks: edit goal, source video role, target material role, edit scope, content to preserve. Then run the same clip through a five-second forward extension with the boundary-frame paragraph written before the new action.

Two generations is enough to feel the difference between editing and re-rolling — and to see why the "content to preserve" block is the one that actually saves the take. For the wider picture of what changed in this release, see what shipped in Seedance 2.5.

Image, video, audio and music models all sit in the same workspace on one balance, so a reference still, an edit pass and an extension are three steps in one place rather than three tools. Start at the create workspace with a clip you nearly threw away.