video-stitchinglisted

# Assemble one coherent H3 Max film H3 Max Reference-to-Video owns normal visual and audio continuity. This skill does **not** use FFmpeg to manufacture it. Read `multi-clip-cinematic-generation` before generating a sequence and `video-editing` for the mechanical media operation. ## Default continuity path For every normal follow-up, use `minimax/h3-max/reference-to-video` and send the accepted immediate predecessor as Video 1. Include the approved character, location, wardrobe/prop, and audio references that matter, then state the intended next camera angle, subject state, action, eyeline, screen direction, light, ambience, and dialogue in the H3 prompt. A changed angle is an intentional cut when the prompt preserves the scene state; it is not a reason to create an FFmpeg repair. H3 returns a new bounded MP4; it does not append bytes to the predecessor. Keep the accepted clips in their recorded order. For each candidate, record its exact path, endpoint, prompt, references, and intended handoff. ## Lightweight clip receipt After a clip downloads, confirm it is usable: inspect its duration and media stream with `ffprobe`, and inspect its opening and ending stable frames. For a successor, compare the predecessor's end and successor's beginning just enough to catch an obvious visual discontinuity. Show the candidate as a preview. Do not create a per-shot FFmpeg seam-preview, hand-pick default trim points, produce a seam-proof document, or run the full final QA suite for e