Result: a 30-second continuous 'oner' made by chaining i2v with last-frame handoff. Full breakdown

by @elenameyer_digital953 in c/seedance-community · · Result

Result: a 30-second continuous 'oner' made by chaining i2v with last-frame handoff. Full breakdown
People said a coherent 30-second continuous-feeling shot was impossible with current models. It kind of is, as a SINGLE generation. But as a CHAIN — each clip's last frame handing off to the next clip's start frame, cuts hidden on motion — I built a 30-second 'oner' that reads as one unbroken take.
youtube.com

People said a coherent 30-second continuous-feeling shot was impossible with current models. It kind of is, as a SINGLE generation. But as a CHAIN — each clip's last frame handing off to the next clip's start frame, cuts hidden on motion — I built a 30-second 'oner' that reads as one unbroken take. It's really eight clips in a trench coat. Full clip + the frame-by-frame handoff breakdown + where the seams are (bet you can't spot them all) in the link.

501 points · 7 comments

Comments

  • @sakurasilva_vfx · · 93 points

    'Eight clips in a trench coat' is the best possible description of how AI 'oners' are actually made and I'm stealing it immediately. The whole art is in the handoffs and the motion-hidden cuts. A real oner is one take; an AI oner is a masterclass in invisible editing. Different skill, equally impressive honestly.

  • @taylorwrightmedia469 · · 57 points

    The invisible-edit oner is arguably HARDER than a real one-take because every seam is a potential failure and you have to engineer the motion to hide each one. Birdman and 1917 faked their 'oners' with hidden cuts too — you're in good company, this is exactly how the pros do it in liveaction.

  • @reesegupta_vfx · · 48 points

    The 'bet you can't spot all the seams' challenge is a power move and I accept. First pass I found maybe 3 of the 7 cuts — the two hidden on the whip-pan and the one behind the foreground pillar got me completely. The pillar wipe especially, that's a classic hidden-cut trick executed perfectly.

  • @miawilliamscreative · · 31 points

    The foreground-object wipe (pass behind a pillar/person/doorframe, cut while the frame is briefly occluded) is THE oner workhorse and nailing it in AI is chef's kiss. When the frame goes dark/blocked for even a few frames you can swap clips invisibly. Engineering those occlusion moments INTO your clips is the real pro skill.

  • @yukinguyen_films2792 · · 33 points

    30 seconds reading as continuous is a genuine milestone for the medium. Two model generations ago this was flatly impossible. The handoff reliability of 2.5's start-frame hold is what makes it work — on older models the handoffs drifted and the illusion broke. Right tool, right technique, right time.