Joins, endings, extension and the video edit

Prefer one generation over a join. Where shots meet, cut between clearly different angles.

• No fixes by hand. A faulty take is regenerated with one diagnosed change, or gets one model edit pass (below); nobody paints, retouches or composites over a take (decided).

• Default joins [proven]: the same anchors and light, joined by real cuts to angles that clearly differ.

• Endings follow the planned cut (decided): the move runs through the last frame for a cut in motion; a settle for a cut after a beat. Always 0.5–1 s of handle at each end, each shot about 1.5–2 times its screen time.

• Planned joins, only where the treatment asks, with a fallback real cut: a match cut (the same lens, size and pose, the move at the same speed); a hidden cut or an empty-frame beat as options. Note it in your take log.

• Every planned join: the original references re-attached, the join hidden in motion where possible, checked for jumps in position, scale and contrast, with a fallback real cut; a jump in geometry or identity means regenerating that segment.

• Extension [official guidance] carries an action past one generation (Template D), never to fix a wrong take.

• The in-between fill (decided): only when the film's core idea is one continuous shot, approved in the treatment and priced before use. Only the gap between two approved ends is generated; nothing outside the gap is touched, and both seams are checked at full size, every frame on either side.

• The video-take edit [best available, not yet validated]: regenerating is the default. One edit per take, for one local fault, saved as a new take and checked at full size on faces and labels. Never chain edits. Regenerate instead when the fault touches the performance or the action, the duration or structure, an object of very different size, new contact or occlusion, or several linked elements.

• The masked local repair. Option: use it when the plan names it for this film [single strong source]. It serves the same case as the video-take edit, one local fault, and regenerating stays the default. It is the same edit, with a black mask over the fault in the source clip and each attached image given one stated job (the correct first frame, the object's design, its later state).

• Video-to-video keeps a clip's motion, with one change stated.

Sources

• Fukushima by @aist (project); @adilinthewildtempo (blog); JOEY, @joeysnoisygroup; Higgsfield's katana workflow.