Prepare references for the current shot, generate several Takes, compare them and adopt the one you want to edit.
Select the current shot
Open Video, select a scene, then a shot in the shot index. The upper-left panel switches between the index and shot information. Check the scene and shot before submitting a job.
The center previews images or video, the lower panel holds prompts and settings, the lower left manages versions, and the right shows history.
Prepare a keyframe
Choose Keyframe as the generation target. The normal mode uses a description and reference images. The narrative mode organizes inputs by location, actors, props and composition. Available inputs depend on the model.
Generated images first enter history. Add the ones you want to versions, then adopt the image you want as this shot’s default reference. Merely viewing or generating an image does not change the adopted keyframe.
You can also use + in the keyframe area to select a local file, project asset or reference tool.
Choose a video generation mode
Switch the target to Video and choose a mode for the task.
| Mode | Purpose | Check before generating |
|---|---|---|
| Keyframe | Animate a reference image, using a first frame or first/last frames as supported | Usable images, camera direction and model requirements |
| Narrative | Describe a performance with character, location, prop and other references | Reference roles, character matching and supported image/video inputs |
| Edit | Modify existing footage | Source video, edit instructions and supported model inputs |
Describe the action and camera changes, choose the model, duration, quality, quantity and aspect ratio, then generate. Models constrain the options; review inputs and settings again after changing models.
Use character voices in video
Bind a voice in Assets first. Rolin checks both the shot’s linked characters and project character images selected as actor references in Narrative mode. Uploading an unrelated image does not create a project character voice binding.
Check the voice row for the current generation and confirm character and voice names. Click a character tag to exclude or restore its voice, or skip voices for this generation without changing project bindings.
A generation currently uses at most three character voices, prioritizing names recognized in dialogue. Multiple samples are trimmed to fit a total of 15 seconds. For larger casts, check which tags are actually enabled and exclude voices not needed in this shot.
When voices are enabled, model choices are filtered to those supporting reference audio. The current integration uses audio-reference-capable Seedance 2.0 models in Narrative mode. In Keyframe or Edit mode, choose between switching to Narrative and skipping voices for this generation. When switching, an available keyframe is added as a composition reference if none exists; it no longer locks the first frame as it did in Keyframe mode.
Audio references still require the images or video expected by the model; a voice sample alone is not sufficient. The Take records the voices actually used. Later changes to character voices do not rewrite old results. Listen to the output: samples guide the sound but do not guarantee an exact match.
Review results and manage versions
Open a history card to play its Take. History groups generation information by batch, with older batches at the top and the latest at the bottom. It initially opens at the top; starting generation scrolls down to the new job. The Chinese interface calls a Take a “候选镜.”
- Add promising results to versions.
- Play retained versions and leave comments as needed.
- Adopt the chosen Take. Each shot has one adopted Take at a time.
Retaining a version and adopting it are different decisions. Deleting a history result also differs from removing it from versions.
Upscale a Take to 2K or 4K
Choose the content first, then increase the resolution of that Take. Upscaling processes existing footage rather than generating a new performance from a prompt.
- Open a successful Take whose video file is already downloaded to this Mac.
- Choose Upscale in the tool panel, or the upscale-to-4K action in a version or history card’s context menu.
- Choose an available upscale model and a 2K or 4K target, then submit.
- Find the new upscale batch at the bottom of history. Play the result when ready; its card has a resolution badge at the lower right.
The result is a new Take and the source remains. If the source was already retained in versions or adopted, the result is also added to versions; otherwise it initially stays in history. It is never automatically adopted.
Upscaling is charged by source-video duration, model rules and target resolution, without adding source and output durations together. Check usage records for the actual charge. The dialog explains billing but does not display a specific advance quote.
Upscaling is unavailable for keyframe images, unsuccessful Takes, files missing from this Mac, or footage with no higher available target. Hover for the reason. A 4K Take cannot be upscaled to the same tier again.
The output frame rate may differ from the source. Inspect picture, sound and frame rate before delivery instead of relying on the 4K badge alone.
Playback and feedback
Open the track strip to inspect frames and audio waveforms, and move the playhead to a specific moment. Safe areas and composition guides help inspect framing; they do not add visible graphics to the generated footage.
To point out a problem, annotate a paused frame and complete the annotation through the comment flow. Some video comment or annotation actions are unavailable when previewing a keyframe image.
Move into editing
Adopted Takes appear in the current episode’s editing library. Initial assembly remains manual: drag or append footage to the timeline. Adopting another ordinary Take does not replace existing clips automatically.
If you adopt an upscaled Take and the current episode’s edit still uses its source Take, Rolin asks whether to replace those clips. Confirming switches their media while preserving in/out points, timeline positions, volume and transitions. Keeping the original leaves the edit unchanged; it does not undo adoption of the upscaled Take.