Wan 3.0 Video Editing

Upload a clip and describe the change. Wan 3.0 video editing rewrites the subject, the background, the lighting or the spoken line while leaving the motion and the framing you already liked exactly where they were.

Three real takes · Press Try this to load the source and the prompt

The regenerated clip: the same walk, restyled

[video editing] · 9:16 · 7s

Street walk

What changes, what does not

What Wan 3.0 video editing changes,
and what it leaves alone

Footage you already have, plus an instruction. What comes back has the thing you named changed, and the rest not.

Wan 3.0 video editing can replace a subject, move the setting, re-grade the look, rewrite a spoken line, remove an object, or continue the story past where it stopped. The action, the framing and the timing carry over because your instruction asked for them, not because the pixels were preserved — an edit is a re-generation that has been told what to respect.

Source clip: ≤ 15 s
Input + output, shared: 30 s
Out, at 30 fps: 2–30 s
Resolution: 480P / 720P / 1080P

Not a timeline editor
Not a masking tool — you describe the change, you do not draw a selection
No file of yours is changed; the output is a new one

One edit, start to finish

What Wan 3.0 preserves,
and one edit start to finish

An edit is only worth doing if the parts you did not mention survive. The things it holds are the things your instruction anchors — and the things it drifts on are the details nobody named.

The source clip: a walk through a street food stall

heldframingwardrobetiming

movedgradesurfaces

Anchored by your instruction, so it holds: camera move and framing, overall pacing and cut points, subject blocking and screen position, the ambient sound bed. Never named, so it drifts: fine background detail, small props at the edge of frame, exact grain and compression texture, incidental sound that was never described.

Here is that rule costing somebody a run. A twelve-second 720P clip — a man in a red jacket walks down a wet street, camera tracking beside him, rain on the soundtrack. The ask was a grey jacket and dawn light. First attempt, *"make the jacket grey and set it at dawn"*, came back with a grey jacket and dawn light — and a different walk, a different street, no rain. The instruction never said to keep them. Second attempt named them: same man, same walking pace, same tracking camera, same street, same rain. Correct. The second prompt is not more creative than the first; it is just longer on the half that says what must not move. Twelve seconds in and twelve out is twenty-four of the thirty-second ceiling, and twenty-four billed seconds.

One thing we will not claim: some sites describe a region-locked edit where every pixel outside the change is byte-identical. That is not in Alibaba's documentation, and the sites making the claim also advertise 4K output that does not exist. Treat this as a very good re-generation that respects your instruction, not as a masked composite. Combining several assets into a scene that never existed is a different job — that is reference to video.

Six kinds of edit

Six kinds of edit,
with the wording that works

Every one of these follows the same shape: name what changes, then name what must not. An instruction that only says what to change invites Wan 3.0 to tidy up other things while it is in there, which is how a simple swap comes back with a different haircut.

Replace the subject
"Replace the man in the red jacket with a woman in a grey coat. Keep the walk, the pace and the street unchanged."
The most common edit and the one that drifts hardest, because a new subject gives the model licence to restage everything around them.
Say this must stay
Walk, pace, street
Ask for this
Who is on screen
Change the environment
"Move this from an office to a rooftop at dusk. Keep the subject, the framing and the camera move exactly as they are."
Naming the framing and the camera is what stops a new setting from arriving with a new shot.
Say this must stay
Subject, framing, camera
Ask for this
Where it happens
Shift the visual style
"Re-grade to a cold, high-contrast look. Do not change the action or the cut points."
Describe only the new look. Anything else you mention becomes fair game to change.
Say this must stay
Action, cut points
Ask for this
Grade, light, texture
Rewrite the dialogue
"Change the spoken line to {We should go}. Same speaker, same timing, same lip position at the end."
Write the line short enough to fit the seconds you asked for, or the lips will not land.
Say this must stay
Speaker, timing, framing
Ask for this
The line, the lip sync
Remove something
"Remove the coffee cup from the desk. Leave everything else in frame untouched."
The shortest instruction on this page, and the one that most needs its second half.
Say this must stay
Everything else in frame
Ask for this
One object, gone
Extend the story
"Continue for eight more seconds: she turns, walks out of frame left, the door closes behind her."
Often the most useful edit of the six. A ten-second clip you already like becomes twenty-two without regenerating the ten you paid for — it continues the motion, the lighting and the audio rather than cutting to something new. Extend twice if you need more, feeding the result back in as the new source.
Say this must stay
Subject, scene, style
Ask for this
What happens next

Dialogue goes in curly braces so Wan 3.0 speaks the line instead of narrating it — the same convention as text to video.

The shared budget

Thirty seconds is a shared budget,
and your input is billed too

This is the Wan 3.0 rule that catches everyone once. Your input clip and its output count against the same thirty-second ceiling: `input seconds + output seconds ≤ 30`.

⚠️ Your source is billed by its own length, on top of the output. A fourteen-second source with a sixteen-second result is thirty billed seconds, not sixteen. Trimming the input before uploading is the single biggest saving available on this page — and a failed edit is refunded automatically, so a bad run costs only the wait.

Shared ceiling: input + output ≤ 30 s; Official
Source clip: ≤ 15 s each · mp4 · mov; Official
Per file: ≤ 100 MB; Official
Frame size: 240 – 4,096 px per side; Official
Aspect: up to 8:1, both directions; Official
Output: 2 – 30 s at 30 fps; Official
Resolution: 480P · 720P · 1080P; Official
Billed: every second in, plus every second outBoth at your output resolution's rate. Audio costs nothing either way.; Official

Upload fourteen seconds and the longest edit Wan 3.0 can return is sixteen. Upload twenty-eight and you have two left. Ask for more than the remainder and the request is rejected before anything generates — so the bar above updates the moment your clip loads and the output slider shrinks to match, which is why the number on the button is always a length Wan 3.0 will accept.
One more rule worth knowing before you hit it. Your source clip is a reference input, which puts an edit in the reference family; first and last frames are the other family, and a request containing both is rejected outright. If you want to pin the opening image of a brand-new clip, that is image to video. Full rates are on pricing.

Audio comes with it

Your clip's soundtrack comes with it

Upload a clip and its audio comes with it, whether or not you meant it to.

This is why a voice you never described turns up in the result, and it carries gender and accent with it. A laugh recorded from a woman still reads as a woman's laugh on a man's face. Say what happens to the sound in the same instruction that says what happens to the picture.

Keep it whole
Name the ambient bed and it survives: "keep the rain on the soundtrack".
Keep the sound
Keep it under new dialogue
Put the new line in curly braces and the bed stays beneath it: "{We should go}, same speaker, same timing".
Rewrite the line
Replace it
Describe the sound you want instead. Anything you leave unmentioned is the part most likely to drift.
New sound

Wan 3.0 video editing is not always the cheap option. Four cases where something else wins: a source that is already near thirty seconds leaves no output budget, so regenerate shorter or trim first; a completely different scene means paying input seconds for footage that survives none of the edit, so use text to video; one exact opening frame is image to video; and the same character across several new shots is reference to video. Editing earns its cost when the motion, the performance or the framing in the source is genuinely worth keeping.

Responsible use

Before you upload footage of someone else

Rewriting a spoken line makes a real person on screen say something they did not say. That is the highest-risk control on this site.

LikenessMaking a real person say words they never said can be unlawful, and it is against our terms without their consent.
VoiceA timbre reference clones how a specific person sounds. Use your own, a licensed one, or a synthetic one.
CopyrightUploading footage or music you do not own transfers no rights to the output. Check the source licence.
ModerationSubmitted media is screened automatically; false positives and negatives happen.
DisclosureFollow the labelling rules of wherever you publish.

Full policy: Responsible use

Questions people actually ask

Wan 3.0 video editing questions

Ten answers · all visible · nothing collapsed

Do I have to sign up for the free edit?

Yes — an email, no card. The free edit runs at 480P on a source clip of five seconds or less. The source cap exists because an edit bills the input as well as the output; the allowance is one edit per account, and it does not reset.

Can Wan 3.0 edit a video I filmed myself?

Yes — the footage does not have to have come out of Wan 3.0 or any other model. Standard mp4 or mov, up to fifteen seconds per clip, and the usual rule that you need the right to use whatever is in it.

How long can my source clip be?

Up to fifteen seconds, and its length comes out of the shared thirty-second budget. Longer source, shorter possible output.

Does it change my original file?

No. Your upload is untouched. What you download is a new video.

Will it change parts I did not mention?

Sometimes, at the edges. Name what must stay the same as explicitly as what changes — that instruction shape is the difference between a clean swap and a redecorated room.

Can I change what someone says?

Yes. Put the new line in curly braces and state that the speaker, the timing and the framing stay put. Wan 3.0 regenerates the lip movement to match.

Does an edit cost less than generating from scratch?

Not necessarily, and often more — because the input seconds are billed too. Editing wins on getting the result you want, not on price.

Can I extend a clip more than once?

Yes. Feed the extended result back in as the new source. Each round is subject to the same thirty-second ceiling, so extend in stages rather than expecting one leap.

Why was my edit rejected before it started?

Almost always the shared budget: input plus requested output exceeded thirty seconds. Occasionally it is a first frame sent alongside the clip, which Wan 3.0 does not allow.

Is this the same as video-to-video style transfer, and can I use the result commercially?

It overlaps but it is broader — style transfer is one of the six edits listed above, and this page also replaces subjects, rewrites dialogue and extends footage. It used to live at `/video-to-video` and now redirects here. On rights: commercial use is included on every paid plan and every credit pack, edits included — publish it, sell it, bill a client for it. Free-tier edits are for evaluation, and Alibaba's own policy for Wan 3.0 applies to every request, so we cannot grant rights the model provider does not. The exact wording is in our terms. A summary, not legal advice.

Bring a clip you already have

Upload it, say what changes and what must not, and let Wan 3.0 hand back the version you wanted. One free edit at 480P on a clip of five seconds or less — an email, no card.

Written and maintained by the wan-3.run editorial teamPublished 25 August 2026Last updated 26 August 2026