Preparing Face and Audio

Section 6 outcome

An approved lip-synced dialogue shot supported by a documented revision process.

Students prepare compatible picture and audio, create lip-synced performances, diagnose artifacts, and use editing strategies when generation cannot meet the shot requirement.

Learning objectives

Explain how to prepare a clear face image and clean dialogue file for reliable lip sync and identify the relevant quality criteria.
Apply the lesson method to the student's own film and produce a usable lip sync input package.
Evaluate the result, document a revision decision, and connect the evidence to the final capstone.

Section 6: Lip Sync and Performance · Lesson 1 of 5 (26 of 45) · 60 minutes · Evidence: Lip sync input package

Warm-Up ~5 min

Watch any film with the sound off and you can still tell when someone is speaking. The jaw drops, the lips close on certain sounds, the eyes follow the thought. Lip sync tools try to recreate that from two files: a picture of a face and a recording of a voice. The tool can only work with what you hand it, so today’s job is to build the best possible handoff.

Production journal: write your answer below.

What must a filmmaker understand before they can prepare a clear face image and clean dialogue file for reliable lip sync?

Instruction ~10 min

What lip sync actually does

Lip sync animates a face so its mouth and surrounding muscles match an audio track. The tool looks at your face image, listens to your audio, and predicts how that face would move while making those sounds. It is a prediction, not a recording, which means weak inputs produce weak predictions. Nearly every disappointing result traces back to the picture or the sound that went in.

Preparing the face image

Check four things before you submit a face:

Face angle: front-facing or only slightly turned. A deeply turned head hides half the mouth and forces the tool to invent it.
Mouth visibility: lips clearly visible, closed or very slightly parted, with no hand, microphone, scarf, or shadow across them. Teeth showing in a wide grin leave the tool nothing to animate from.
Neutral expression: a calm, relaxed face is the best starting point. The tool can add movement to a neutral face; it struggles to undo a frozen laugh.
One face: a single face, large enough in the frame to see the lips and eyes. If a crowd is in the shot, the tool may animate the wrong person.

Preparing the audio

Clean dialogue means one voice, speaking clearly, with no music, wind, or room echo underneath. A tool listens to every sound in the file, so a gull cry or a drum loop can look like a mouth movement. Then check three more things:

Trim: cut dead air at the start and end, but keep a fraction of a second of silence so the face can settle in and out of the line.
Length: many tools limit how long a clip can be, and long clips drift more. Plan on short lines, roughly five to ten seconds, and check your school-approved tool’s limit.
Level: speech should be steady and audible, not clipped or whisper-quiet.

Worked example

For Last Light, the student wants Mara to say, “The lamp has never failed us.” The portrait she likes best has Mara glancing sideways at the sea with her scarf pulled up over her chin. Rejected: the mouth is hidden and the head is turned. A second portrait has her facing the camera, lips together, round copper glasses clear of her cheeks, soft light: chosen. The recording of the line has two seconds of wind before the first word and a gull at the end. The student re-records it indoors, trims it to a 6.4-second clip with a quarter second of quiet on each side, and names it with the course convention.

Common mistakes

Choosing the most dramatic portrait instead of the most readable mouth.
Feeding in a recording with music or sound effects already mixed in.
Submitting a thirty-second speech when one sentence is all the shot needs.
Using a face or a voice you do not have the right to use.

Workflow Study ~10 min

Creators who make talking characters follow a quiet preparation routine before they ever press generate. Study it as if you were looking over their shoulder, and for each step ask: who decides, the person or the tool?

The creator starts with a character image and checks the face the way an editor checks a shot: is the mouth visible, is the head square to the camera, is anyone else in the frame? If not, they make a new image or crop one rather than hoping the tool will cope. Then they turn to the voice. They either record a line or produce one with an approved voice tool, listen on headphones, and cut away everything that is not speech. Only then do they pair the two files and start the tool.

Notice what the software does and does not do. It may offer to detect a face or to extend a clip, but it will not tell you that the audio has a hum in it or that the portrait hides half the lips. Those judgments are yours, and they are cheap to make now and expensive to fix after generation.

As you study, record each decision point in your production journal.

Decision point	What the filmmaker does	What the tool automates	What still needs human review
Choose the face image	Checks angle, mouth visibility, expression, and number of faces	Can generate or detect a face on request	Whether the lips are truly readable at full size, not just in a thumbnail
Prepare the audio	Records or generates one clean line, then trims silence and noise	May denoise or normalize volume if asked	Whether anything but speech remains, and whether the voice is authorized
Set clip length	Cuts the line to the length the shot needs and the tool allows	Enforces a maximum length or rejects the file	Whether the line still makes sense when shortened
Label and package inputs	Names files, records source and permission	Nothing; this is a human record	That the face and voice are both documented as authorized

Guided Studio ~15 min

Studio task: Evaluate face angle mouth visibility audio length and silence in sample inputs.

Your teacher provides a set of sample faces and sample audio clips, some good and some flawed on purpose. Work with a partner.

Look at each face image at full size. For each, score face angle, mouth visibility, expression, and face count from 0 to 2. Write one phrase naming the biggest weakness.
Listen to each audio clip on headphones. Write the length in seconds, how much silence is at the start and end, and whether you hear anything besides one voice.
Pair the best face with the best clip and write a one-sentence reason for the pairing.
Choose the single worst input in the set and explain which failure it would cause (for example, a hidden mouth leading to an invented one).
Compare your scores with another pair and discuss any score that differs by two or more.

Coaching tips. View the face image at the size it will appear in your film, not zoomed in on a single feature. If you cannot tell whether the lips are parted or closed, the tool will not be able to either. For audio, close your eyes and listen once for noise only; the second time, listen for words.

Input check sheet (copy into your production journal)

FACE IMAGE (file name): ______
Angle 0-2: __  Mouth visible 0-2: __  Neutral expression 0-2: __  One face 0-2: __
Biggest weakness: ______

AUDIO (file name): ______
Length (seconds): ____  Silence at start/end: ____ / ____
Anything besides one voice? ______

Source and permission for face: ______
Source and permission for voice: ______

Knowledge check

+50 XP

Which face image is the best input for lip sync?

Myth or Fact?

Background music already mixed into the dialogue file helps the tool time the mouth.

Fill in the blank

A recording of one clear voice with no music or wind underneath is called clean .

+50 XP

A dialogue clip is twenty-five seconds long and the tool drifts badly at the end. What is the best fix?

Myth or Fact?

It is acceptable to use a recording of a famous actor's voice if you only use it in class.

Fill in the blank

Keep a fraction of a second of at the start and end of a trimmed clip so the face can settle.

Describe one input condition in your own film (a face or a recording) that you expect could cause a lip sync failure and how you will prevent it.

1 / 7

Independent Studio ~15 min

Studio task: Prepare and label the approved face and dialogue inputs.

Prepare the inputs for the first dialogue shot of your own film.

Choose or create a face for the speaking character. Create at least three candidates and pick one using the input check sheet.
Write or choose the line for the shot. Keep it to a single sentence or two, about five to ten seconds when spoken.
Record or generate the audio with an approved method, then trim and clean it. Listen on headphones.
Label both files with the course naming convention and write one line of provenance for each: who created it, with what, and who authorized it.
Record one revision decision, such as replacing a portrait or re-recording a noisy take, and why.

Checklist before you submit:

One face, mouth clearly visible, neutral expression.
One voice, no music or noise, trimmed, short enough for the tool’s limit.
Provenance and permission written for both files.
If you cannot use a generation tool, your teacher-approved alternative is documented.

Exit Evidence ~5 min

Name one input condition that would cause failure.

Student Deliverable & Assessment

Deliverable: Lip sync input package.

Submit the lip sync input package. The file must use the course naming convention and include enough process evidence to show the student's decisions and revisions.

File name (course naming convention): AIF_S06-L01_lip-sync-input-package_v01_YOURINITIALS

Criterion	Advanced	Proficient	Developing	Beginning
The work addresses the stated story or production purpose.	The face and line serve the planned shot, and the choice is justified against the story moment.	The face and line suit the planned shot.	The face or line only loosely fits the planned shot.	The inputs do not connect to a planned shot.
The student applies the required method accurately and produces usable capstone evidence.	Both inputs meet every check, with measured length and trim noted; another student could reproduce the package.	Both inputs meet the checks and are labeled with the naming convention.	One input meets the checks; the other has an unresolved flaw.	Inputs are unsuitable or unlabeled.
The student identifies at least one quality issue and records a meaningful revision decision.	Names the weakest input, tests or compares an alternative, and records a specific revision.	Identifies a quality issue and records a meaningful revision decision.	Notes a problem without a clear revision.	No revision decision recorded.
The work complies with consent rights attribution privacy disclosure and school policy.	Provenance and written authorization recorded for the face and the voice, with an alternative route documented.	Provenance and authorization recorded for both files.	Provenance is recorded for only one file.	Uses a real person’s face or voice, or unlicensed material, without permission.

Responsible Production

Requirement: Use only an authorized character face and voice.

A face and a voice are a person’s identity, and lip sync joins them. Use only a face you created, generated with your school-approved tool, or that a consenting person released in writing, and do the same for the voice. Never pair a real person’s likeness with words they did not say. Write down where each file came from before you generate, and keep that note with the files.

Up NextComplete & Continue →