Point and Edit: How MiniMax H3 Changes Video Post-Production

Post-production has always been organised around a hard fact: changing something in a finished shot is expensive in proportion to how much of the frame it touches.

Remove a logo from a wall and someone rotoscopes it, patches the plate, matches the grain, and tracks it across every frame. Change a line of dialogue and you book the actor for ADR, then spend an afternoon making a new performance sit inside an old mouth. Swap the product in a hero shot and you’re in comp for a day, or back on set.

None of that is inefficiency. It’s the actual cost of precision — of altering one element while every other element stays exactly as it was, frame after frame. Post exists because that’s hard.

Generative video, for most of its short life, offered nothing here. It could make new footage. It could not modify yours.

What “point and edit” actually means

The capability that changes this is narrow and specific: local editing on existing footage.

Minimax H3 modifies a designated element while holding the rest of the frame stable. In practice that covers most of what a revision request consists of — replacing, adding, or removing objects and people; changing backgrounds, environments, and lighting; adding or adjusting visual effects; modifying a character’s motion or performance; replacing dialogue with the mouth re-forming around the new words; cloning or transferring vocal timbre.

The operative word is holding. A model that regenerates the scene with your change applied has done something categorically different from editing. Post doesn’t need a new video that resembles the old one; it needs the old one with a correction in it. That’s why H3’s position at the top of the Artificial Analysis video editing leaderboard, ahead of Seedance 2.0, is the relevant benchmark for this discipline — it measures preservation, not just fidelity.

Where it lands in an existing pipeline

Not as a replacement for a finishing suite. As a front-end for decisions.

Consider the dialogue case, since it’s the clearest. Conventionally, changing a line means ADR, then lip-sync repair, then re-mixing against the original ambience. Three specialists, a scheduling problem, and a cost the client has to approve before anyone knows whether the new line is even better.

With H3, the version exists in minutes. Because audio is generated natively alongside the picture — dialogue, ambience, foley, and music in the same pass — the alternate line arrives with sync and with sound around it, not as a silent mouth waiting for a mix. The client hears the option, picks one, and only then does the real ADR session get booked, with a known target.

That’s the pattern generally. The value isn’t that generated output replaces finished work; it’s that decisions that used to require finished work can be made before it exists.

The cleanup case

The other obvious application is unglamorous and probably more valuable: removals and replacements.

Every project accumulates them. A brand mark that legal flagged. A crew member reflected in a window. A product that changed packaging between the shoot and the launch. A background element that reads differently in one market.

These are pure preservation problems — the whole frame must stay identical except for one region, across the full duration. Historically that’s roto, paint, and tracking. Local editing addresses a meaningful share of them directly, particularly on shots without complex occlusion or heavy motion.

Not all of them. Which brings us to the part that matters most for anyone considering this seriously.

The limits are real and worth stating

If you work in post, you’ll hit these within a day.

Resolution and delivery. Output tops out at 1440p, at 24 FPS. That’s genuinely fine for social, e-commerce, and much broadcast-adjacent work, and it’s below what a 4K deliverable requires. For finishing on a high-end campaign, this is a previz and options tool, not a mastering tool.

No passes, no alpha, no scene data. You get a picture. You don’t get mattes, depth, or separated elements to comp with. That means output integrates as a flat clip, not as material a compositor can take apart and rebuild.

Duration. Fifteen seconds is the ceiling. Many shots fit inside that; sequences don’t.

Edit magnitude. Preservation is strong for contained changes and degrades as the change grows. Replacing a central subject or swapping an entire environment disturbs far more of the frame than removing a sign from a wall. There’s a threshold past which you’re effectively regenerating with extra steps, and finding where it sits for your material takes actual use rather than a benchmark.

Frame-exact control. You’re directing a model, not setting keyframes. For work where a change has to land on a specific frame and match a specific curve, conventional tools still win.

Why it’s worth having anyway

Because most of what post costs isn’t the final execution. It’s the iterations before it, and the decisions made without enough information.

The expensive pattern in any post schedule is discovering late that an approved direction doesn’t work — after the roto is done, after the mix is locked, after the client has seen a version they now want changed. Anything that lets a team see and reject options earlier removes cost that never appears as a line item.

Per-second generation cost sits well below comparable models, which matters here for the same reason it matters everywhere: the number of options you can afford to produce and discard determines the quality of the one you keep. Teams evaluating minimax h3 ai for post work should test it on their own real cleanup shots rather than on demo prompts — that’s what reveals where the threshold sits.

The honest framing

This doesn’t replace a post house. It doesn’t produce deliverables for a high-end campaign, and it doesn’t give a compositor anything to composite with.

What it does is move a category of question — what if we removed that, changed that, said it differently — from the expensive end of the schedule to the cheap end. In a discipline where most of the money is spent finding out what the client wanted, that turns out to be a substantial change.