How to change one expression in a finished animation

Last updated 2026-09-22 · 8 min read

What is the short version?

The animation is finished and exactly one thing is wrong with it: the character's mouth is still moving after the line has ended. Regenerating the film would fix that and change nine other things you liked, which is why Vizard Agent edits rather than rebuilds.

1. Tell Vizard Agent the characters, movement and timing all stay.
2. Say which seconds contain the fault and what it should be instead.
3. Ask to see the face compared with the original before delivery.

What do you need before you start?

The finished video and a precise description of the new expression. Vizard Agent works from what you say to the frame, so "mouth closed, not speaking, smiling warmly" gives it a target where "make him look happier" does not.

• The finished video. The version you approved.
• The moment. A timestamp or "the last three seconds".
• The new expression. Described concretely.
• What must not change. Characters, movement, timing, style.
• The output length. Usually identical to the source.

What do you type into Vizard Agent?

Lead with what stays rather than what changes. The sentence "keep the exact same animation, characters and movements" does more work than the instruction that follows it, because it tells Vizard Agent this is a correction to an approved film rather than another attempt at one.

Prompt

Variants worth knowing:

• "Keep the exact same animation." Correction, not regeneration.
• "Mouth closed, not speaking, smiling warmly." A describable target.
• "Same length." Stops the ending being retimed.

What does Vizard Agent actually do?

Here is the Vizard Agent sequence on an animated short whose closing expression was wrong. Almost every step is verification, because the risk in this job is not failing to change the face — it is changing everything around it.

1. Inspects the uploaded video and extracts the ending frames.
2. Views the final three seconds to see exactly what the face is doing.
3. Checks the analysis capability available for the mouth movement.
4. Inspects the boy's final mouth movement specifically.
5. Checks the editing options and their precise controls.
6. Prepares the short final segment for a targeted edit.
7. Tests the facial-expression edit on that segment alone.
8. Reviews the test render for visual drift against the untouched tail.
9. Compares the final face at full size, then locates the face in both versions.
10. Takes a tighter comparison of the mouth specifically.
11. Renders the approved edit and joins the untouched opening back on.
12. Corrects the final duration with exact trims, then checks the frames and the waveform.

Step six is what keeps the film intact. Only the last few seconds are sent through the edit, so the rest of the animation is the original file rather than a re-render that happens to look similar.

Steps nine and ten are the check that matters. A generative edit can change the face and also nudge the character's colour, the background's shade or the line of the shoulder, and none of that is visible until you put the two frames side by side at full size.

Step twelve is the detail that makes it deliverable. Joining an edited tail to an untouched opening rarely produces exactly the original duration, and a film that is nine frames longer will not drop into a timeline that expects the old one.

What does the result look like?

The same animation with one expression different in the closing seconds: the same characters, the same movement, the same colours and the same runtime as before, and a face that now does what you asked Vizard Agent for.

When does this not work well?

Some expressions cannot be changed in isolation. If the mouth movement is tied to audio that continues, if the character turns during the moment, or if the style is highly detailed, the edited face will not match the frames either side of it.

• Audio that continues. A closed mouth over speech looks wrong.
• Movement during the moment. The edit has to track as well as replace.
• Highly detailed styles. Small differences read as a different character.
• Very short moments. Not enough frames to blend.
• Faces in the far background. Too few pixels to edit convincingly.

How do you fix a result that came back wrong?

Compare the two versions and name the difference you can see. Vizard Agent can rebuild the segment against a tighter constraint, and a note as specific as "his jaw is a different shape" is far more actionable than saying it does not look like him.

• "The face drifted." The segment is re-edited with a tighter comparison.
• "The colours changed." The edited tail is graded back to the original.
• "The join is visible." The boundary is moved to a frame with less movement.
• "It is longer than the original." The duration is trimmed to match exactly.

How does Vizard Agent compare to doing it yourself?

By hand, with the project file, this is a small fix. Without the project file — which is the usual case when the animation came from somewhere else — it means rebuilding the shot, and most people simply live with the mouth still moving.

Common questions

Do I need the project file? No. Vizard Agent works from the finished video.

Can it change an expression in the middle? Yes. Name the timestamp and Vizard Agent edits only that segment.

Will the character still look the same? That is what the full-size comparisons are for, and you should ask to see them.

Can it close a mouth that is speaking? Yes, provided the audio does not continue under it.

What about the eyes? Same approach — describe the expression precisely and Vizard Agent works to it.

Can it change two characters at once? Yes, though Vizard Agent will check each face separately.

Does the whole video get re-rendered? No. Vizard Agent joins the edited segment to the untouched original.

Will the join show? Not if it lands on a still moment, which is what Vizard Agent looks for.

Can it keep the same length exactly? Yes. Vizard Agent trims the result to the source duration.

What if the style is very detailed? Then the edited face is harder to match, and Vizard Agent will say so.

Does it work on live action? Yes, though expressions on real faces are less forgiving.

Can I see the two versions side by side? Yes. Ask Vizard Agent for the comparison frames.

What if I want the line removed too? Then the audio changes as well, and Vizard Agent treats it as two edits.

How long does this take? Most of the time goes on checking, not on the edit itself.