Seedance 2.5 Prompting Guide (Part 2): Video Editing, Extension, and Final Assembly
Learn Seedance 2.5 prompts for instruction-based editing, reference-image editing, dialogue localization, extension, automatic assembly, and seamless transitions.
An editing prompt should read like a precise change request: identify the target, the scope of the change, the properties that must remain intact, and the time range in which the edit applies. This reduces unintended changes to identity, camera, timing, and sound.
If you have not yet defined asset roles and a timeline, begin with Part 1: Prompt Structure and Reference Generation.
Rules for Editing Tasks
Editing, first/last-frame generation, and extension are locked tasks. The input is not an ordinary style reference; it defines a timeline or boundary frame that the output must fit.
| Task | Locked property | Prompt and output guidance |
|---|---|---|
| Video editing | Aspect ratio follows the source; duration stays close to the original | Use adaptive ratio and -1 duration; explicitly say add, remove, replace, or modify |
| First / last frame | Aspect ratio follows the first frame; the last frame should match it | Mark images as first_frame / last_frame; duration can be set independently |
| Video extension | Aspect ratio follows the source clip | State forward or backward and the number of seconds; MOV input and output usually preserve audio-visual continuity better |
When uploading more than one video, begin with “Edit Video 1” or “Extend Video 2.” Do not make the model infer the target from context.
Write the Prompt as an Edit List
A reliable editing prompt has three parts:
- Lock the target: Which clip, person, region, or time range should change?
- Define the change: What becomes what, and is the transition instant, gradual, or continuous?
- Protect everything else: Which actions, camera moves, timing, lighting, lip movement, or sounds must stay unchanged?
Edit seconds 4–7 of Video 1. Replace only the dark jacket on the person at right with the ivory trench coat from Image 1; its material and folds must respond naturally to the original movement. Preserve identity, expression, body movement, background, camera, lighting, edit rhythm, and original sound. Keep everything before second 4 and after second 7 exactly as in the source.Instruction-Based Video Editing
Without a reference image, describe the visible target clearly and list the original properties that must be protected.
Prompt
Edit Video 1. Preserve the original composition, camera, lighting, background, and performance timing. Change only the woman's age and expression.
Let her age naturally from her twenties to her sixties. Her facial structure and identity remain consistent while skin, hair color, and facial detail change continuously, without jumps or flicker. The restraint in her eyes gradually releases, a tear travels from the corner of one eye, and the corners of her mouth slowly rise until she smiles through tears.
Keep the entire shot continuous. Do not change body movement, depth of field, color grade, or sound.Reference-Image Editing
Map each reference image to a specific object in the source video. When movement comes from the source, say “preserve the original action order and timing exactly” instead of describing the action again and risking drift.
Prompt
Edit Video 1. Preserve both performers' action order, speed, distance, camera, and original edit rhythm exactly.
Replace the setting with the medieval fortress platform from Image 1: flat stone paving, castle walls, distant mountains, wind, and light mist. Replace the clothing of the dark-clothed performer with the outfit in Image 2, and the clothing of the light-clothed performer with the outfit in Image 3. Keep both identities and faces unchanged.
Add only restrained environmental detail: wind moving the garments, small dust puffs at contact points, cool metallic highlights, subtle film grain, and an epic color grade. No exaggerated magic effects. Synchronize the background score to the original action beats.
Audio Editing and Dialogue Localization
Separate voice, lip synchronization, and visual protection rules so translating dialogue does not also alter the shot or performance.
Prompt
Edit the dialogue in Video 1. Translate all speech into natural Chinese and precisely regenerate lip movements for Chinese pronunciation. Preserve the speaker's vocal character, emotional intensity, pauses, and natural speaking rhythm.
Do not add subtitles. Keep the background music, ambience, sound effects, camera, character appearance, action, visual rhythm, and duration unchanged.Video Extension
Begin by stating whether to extend forward or backward and by how many seconds. Continue from the source clip's last frame—subject position, movement direction, lighting, focus, and sound—rather than starting a new scene.
The extension may have a small volume difference. Using a Seedance 2.5 source clip and MOV for both input and output generally produces a smoother join.
Prompt
Extend Video 1 forward by 5 seconds after its ending. Preserve the same flower, natural light, macro depth of field, color, and continuous ambience.
A bee flies in and lands on the flower. Cut closer to a macro view of golden pollen collecting on its legs and abdomen. It lifts off and the camera follows to a nearby flower of the same species. In slow motion, pollen shakes loose from its hairs and falls onto the stigma, clearly magnifying the instant of pollination.
Do not reintroduce the setting or alter the flower's state in the original last frame. Output MOV.Automatic Video Assembly
For a multi-image edit, do not simply say “make a video.” Specify the format, visual packaging, allowed image motion, music direction, and the source details that must not change.
Prompt
Arrange Images 1–8 freely into a vertical café check-in vlog about the same dog posing in different cute outfits.
Use hand-drawn animated doodles and light cutout collage as the visual package. Add a restrained number of sketched lines, stickers, and rhythmic transitions for a relaxed social-media feel. Images may have subtle Live Photo-style movement and natural push-ins or pull-outs, but do not alter the dog's identity, clothing details, café furnishings, or original composition.
Generate lively but not overwhelming background music, and synchronize transitions to the beat.
Seamless Video Transition
Describe how Video 1 exits, what transformation occurs during the bridge, and from which viewpoint Video 2 begins. “Connect naturally” is usually not specific enough.
Prompt
Connect Video 1 and Video 2 seamlessly while preserving the content, speed, and color of both source clips.
Continue from the end of Video 1. The camera races toward the highest point, pauses briefly, then immediately reverses and dives straight down. During the dive, the black-and-white dominoes stretch upward and gradually transform into city towers while the tabletop texture and surroundings become the urban space in Video 2. Camera direction, speed, and motion blur remain continuous until the movement enters Video 2's opening viewpoint naturally.
No black frame, white flash, or visible hard cut. Keep the subject transformation continuous.Editing Preflight Checklist
- Does the first line name the video to edit or extend?
- Are the target, time range, and transition from A to B explicit?
- Are all protected actions, camera moves, timing, visuals, and sounds listed?
- Is every reference image mapped to a specific object in the source video?
- Are the extension direction, duration, and source ending state clear?
- Does a seamless transition include exit, bridge, and entry stages?
- For video extension, do the input and output both use MOV?
Back to Seedance 2.5 Prompting Guide (Part 1): Prompt Structure and Reference Generation.
Seedance 2.5 Prompting Guide (Part 1): Prompt Structure and Reference Generation
Learn Seedance 2.5 prompt structure through task selection, asset mapping, continuous timelines, blockouts, storyboards, and keyframes.
Seedance 2.0 Complete Prompting Guide
Learn how to write effective prompts for Seedance 2.0 text-to-video, image-to-video, and reference-to-video workflows.