Advanced Workflows

Advanced workflows

Continue an existing video

Use Existing Video Context when scene 1 must follow an imported clip.

Plan ─────────────────────────┐
Native VIDEO ─────────────────┼→ Existing Video Context → Loop Start
or decoded IMAGE/AUDIO + FPS ─┘

Use one source route:

• native VIDEO preserves exact FPS and may expose embedded audio;
• decoded IMAGE accepts optional AUDIO and requires the real source FPS.

The node normalizes the source to the Plan canvas and 24 fps, uses its final context frames as scene 1's predecessor, and can prepend the normalized original during assembly. The imported tail is fingerprinted for safe resume.

Start with Guide, 22 context frames, video encode, head anchor, and Loop Trim match_tail=true. Use a masked AV extension workflow when the imported picture and sound must be protected inside the target latent.

Long visual context

56 is a valid advanced Guide length. It carries about 2.33 seconds of motion in 17 video-latent steps, but head mode regenerates and removes all 56 frames from every continued scene. Start with 22 and increase only when a long camera move or performance genuinely needs more history.

AV-mask modes do not accept arbitrary lengths. They require a shared 24 fps video / 40 Hz audio boundary: 39, 90, 141, 192, or 243 frames.

Last-frame destinations

When core H3 Image to Video supplies last_frame, Contex Loop preserves that target on continuation scenes. A repeated first-frame anchor cannot occupy the same opening coordinates, so the incoming context takes priority there.

For alternating destinations, drive Frame Index Switch with Current Shot's scene index and connect it to Frame Gate.last_frame.

Re-film a synchronized performance

Use a video reference when one performance supplies action, cadence, and timing across new viewpoints or appearances. Tagged Motion Ref is preferred when the clip should transfer performance without importing identity, wardrobe, setting, lighting, or composition.

The legacy Reference Video Prep route converts native VIDEO or decoded IMAGE/AUDIO to exact 24 fps Ref2VA input and rejects sources shorter than the requested H3-valid length instead of padding or stretching them.

External stitchers and visual blends

Loop Trim's images_with_overlap output retains part of the repeated picture context separately from the normal hard-trimmed output. Connect Current Shot's video_blend_frames to retain_overlap_frames.

Scene N controls the boundary entering scene N. 0 remains a hard cut. The blend setting affects assembly only; it does not change diffusion, clean scene frames, or audio.

Source-aligned performance timelines

Source Timeline is the preferred route for long video/audio material. Current Shot requests one exact overlap-aware window, and Tagged Motion Ref can advance on the delivered timeline so a protected continuation prefix does not replay the same motion interval.

See References and source media and Continuity and audio.