1An avatar represents someone in a scene

Avatar is a visual representation used as a presenter or character.

It can start from a photo, a recorded person, or an invented character.

A digital twin uses material from a real person.

A fictional character needs a consistent appearance but does not represent a lived experience.

At Oficina Clara, Otávio wants to explain a technique using his own identity.

Bia is also planning Nina, a fictional presenter.

These paths should not be mixed up.

An invented character can present information, but should not be given a real story they never lived.

Bia, a 32-year-old producer, keeps Nina separate from the teacher’s identity.

2Three inputs, three review criteria

The face sets the appearance, the voice provides speech, and movement guides the performance.

A feature may combine these parts, but each still needs review.

In HeyGen, III, IV, and V label system generations; your account shows which fit your avatar type.

Do not choose based only on the higher number.

Check the required input and displayed cost.

A path that accepts a photo does not necessarily accept a full performance reference.

Record the feature used so you can repeat the test later.

Limit = cost per attempt times the maximum number of attempts.

Fictional example: 10 credits per attempt, up to three attempts, for a 30-credit limit.

Otávio, a 59-year-old instructor, chooses to teach a technique using his own image.

3Check the decision against its use

Photo, voice, and movement are different inputs; one reference does not automatically replace the others.

Otávio uses the sheet to check: inputs: face, voice, and movement kept separate.

1Set up light, framing, and camera support first

Place your phone on a tripod or firm support, with the lens near eye level.

Use soft light in front of you or slightly to one side.

Avoid a bright window behind your head.

Clean the lens, look at it, and keep your mouth visible.

Make a short test before the main recording.

For a waist-up shot, keep your head and hands in frame during the gesture.

For a full-body shot, include your feet too.

The photo you use later should match this framing.

Stay in the same place during the first test.

Bia checks the test before asking for a longer recording.

2The file must fit the path you chose

A digital twin recording may require speech and expression.

A motion reference needs one continuous shot and a readable gesture.

Check the feature requirements before the final recording.

Do not assume one file works for every mode.

In HeyGen, creating a digital twin includes confirmation by the person represented.

Follow the text and process shown by the platform.

An external written permission does not replace this step.

Keep original files separate from edited versions.

Otávio realizes he needs room above his hand when he waves.

3Check the decision against its use

The test video can be used as final material only if it meets the requirements shown by the tool.

Otávio uses the sheet to check: requirements: check the chosen feature.

1One line, one purpose

Choose one purpose: introduce, explain a step, answer a question, or invite people to continue.

Use short sentences with words you would really say.

Write numbers the way you want them spoken.

Read names aloud to catch unclear pronunciations.

For Oficina Clara: Today we’ll choose a paintbrush.

Look at the shape of the tip before you begin.

This line guides one simple action.

It does not need to tell the workshop’s story or promise a result.

Bia writes performance directions outside the text the voice will read.

2The pause is part of the message

Record in your phone’s Voice Recorder app. Listen and check the time shown in the recording app.

If it runs longer than fifteen seconds, cut one idea.

If it is too short, leave a natural pause instead of filling it with adjectives.

Keep spoken text separate from performance directions; those belong in another field.

A small smile is a performance direction, not a phrase for the avatar to say.

If you put directions in the spoken text, the voice may read them aloud.

Review the audio before spending on animation.

Otávio records a rehearsal to find where he needs to breathe.

3Check the decision against its use

Listen to the file before animating; written preview text does not confirm pronunciation.

Otávio uses the sheet to check: rehearsal: measured audio duration.

1Text to speech turns writing into audio

Text to speech reads a message using a selected voice.

In ElevenLabs, open Text to Speech, paste your line, and choose a voice and model that support Portuguese.

Check the cost before generating.

Listen to the whole output and save the approved file.

Choose a voice that fits the purpose.

For teaching a technique, clarity and pauses help more than exaggerated acting.

Do not assume a Brazilian accent based on the page language.

Listen for names, numbers, and words from your topic.

Bia tests the workshop name before animating a whole series.

2Review in three passes

First check the words: is anything missing or added? Is the name pronounced correctly?

Then assess pacing and pauses.

Finally, listen on your phone at a comfortable volume to catch noise and cuts.

Note the section to change before generating again.

Keep the text, voice, and settings with the audio.

If the workshop has access to INEMAVOX, it may be a supervised alternative.

Confirm access and the saved voice with the person responsible.

The course does not assume this local service is available online.

Otávio prefers clear speech to a delivery that rushes.

3Check the decision against its use

Listen to words and pauses before judging the face. Fix audio problems at their source.

Otávio uses the sheet to check: review: words, pacing, and final file.

1A voice clone approximates vocal traits

Cloning uses a recording to approximate tone, rhythm, and pronunciation.

Record your own voice in a quiet place, with no music or other people talking.

Keep a steady distance from the microphone and avoid clipped sounds.

Listen carefully before you submit it.

If the voice crackles or sounds harsh on louder parts, move the microphone a little farther away or speak more softly.

Record it again and compare.

On ElevenLabs, instant voice cloning lets you record or upload audio and confirm your rights to the voice.

Professional voice cloning has a separate verification process for the voice owner.

Do not treat the two processes as equivalent.

For this exercise, use only your own voice.

Bia asks for a new sentence to compare the clone with the reference.

2Check the resemblance without confusing accent with identity

After creating a clone, use a sentence that is different from the reference recording.

Check whether the voice sounds recognizable, clear, and natural.

A change in accent or emotion can reduce the resemblance.

Keep the original audio so you can make a direct comparison.

Other services, such as MiniMax, also offer voice cloning.

Check that service’s own requirements and controls; steps from one service do not automatically apply to another.

The simplest alternative is still to record your own narration, when that works for the video.

Otávio records his own voice and checks that the last word was not cut off.

3Check the decision against its use

A recorded sample, a created clone, and synthesized speech are three different deliverables.

Otávio uses the worksheet to check: comparison: a new sentence, not the reference.

1Choose the avatar in the right feature

In HeyGen, use Avatars to create or select your presenter.

For a photo, follow the Photo Avatar steps; for a video-based version of yourself, follow Clone a Real Person and complete the required confirmation.

Check the visual style and compatible systems in your current account.

HeyGen guides describe different steps for different features.

Some describe Single Scene; others, AI Studio.

Choose an available presentation mode and record its name.

Do not assume every avatar supports the same system: III, IV, or V.

Bia confirms the avatar selection before approving the cost.

2Approve the voice before generating

In the presentation project, select the avatar and upload the reviewed audio using the audio-file option, when available.

If you use the service’s text and voice, listen to the preview.

Choose a format and simple performance.

Check the duration and cost before generating.

When the file is ready, watch it all the way through.

Compare the words, mouth, eyes, and hands.

An audio preview does not confirm the final movement.

Download the approved version and save the settings.

If the face distorts or the speech gets cut off, note where it happens before trying again.

Otávio watches the entire file to check mouth movement and pronunciation.

3Check the decision against its use

An audio preview and a still image do not prove the quality of an animated avatar.

Otávio uses the worksheet to check: output: video watched and saved.

1Clearly mark the spoken text

Separate the scene, exact speech, and performance direction.

For a presenter facing forward, ask for a still camera, a visible mouth, and one short sentence in Portuguese.

Use quotation marks to mark what will be said, when the mode accepts speech in a prompt.

This makes review easier, but it does not force the generator to reproduce every word.

The course offers five purposes: greeting, tip, explanation, answer, and invitation.

Choose one for each test.

Asking for all of them in a short generation mixes goals and makes it harder to find what failed.

Check the audio that was produced; do not assume it follows the text.

Bia uses one sentence per attempt to pinpoint the error.

2Choose the tool based on the control you need

Kling offers options for native audio, avatars, and later lip sync, depending on the mode and account.

Grok Imagine can also create video with audio in compatible settings.

A feature documented for programmatic integration may differ from the app.

Check what your screen actually accepts.

When you need to use an approved voice file, choose an option that accepts that audio, such as a compatible avatar mode.

If the speech is only meant to explore an idea, a native-generation test may work.

Do not present an exploratory attempt as final narration.

Otávio compares the audio word by word before approving the test.

3Check the decision against its use

Only listening and comparing the output lets you say whether the text was spoken correctly.

Otávio uses the worksheet to check: review: compare the words in the audio.

1Use an image and audio to make an avatar

In Kling Avatar, select an image of yourself or your character, then add the approved audio.

The Avatar 2.0 guide also lets you direct expression and performance.

Start with a small gesture and an unobstructed face.

Check the cost and make a short attempt.

This option starts with an image and audio.

The separate Lip Sync feature can start with an existing video and adjust the mouth to the speech.

If your account offers this second option, also provide a clip with the face visible.

Record which operation you used.

Bia separates an audio pronunciation error from a facial movement error.

2Listen, watch, then combine

Listen without watching to check the words.

Then look at the face, teeth, and edges of the mouth.

Finally, follow the sound and image together.

Look for noticeable delays, a cut-off sentence, or a mouth moving in silence.

Record the moment when the problem occurs.

A line with a clear pause makes this check easier.

Do not expect one fixed mouth shape for each letter: movement depends on neighboring sounds.

The first goal is to spot clear mismatches and decide whether the piece can be used.

Otávio marks the moment when the mouth keeps moving after the last word.

3Check the decision against its use

A sequence of photos cannot show sync; you need to watch the video with its audio.

Otávio uses the worksheet to check: review: words, mouth, and pauses.

1One person, one gesture, steady camera

Record yourself making a short wave, with your phone on a tripod.

Keep one person in view, with no cuts or camera movement.

Use a moderate speed.

Leave room for the hand throughout the gesture.

Choose a photo of yourself or an invented character with compatible framing.

A medium shot matches a medium shot; a full-body shot matches a full-body shot.

The photo also needs room for the action.

A hand resting under the chin does not show that the frame can fit a high wave.

Plan that margin before uploading.

Bia tests a small gesture before asking for a performance that moves around.

2Check the gesture and identity separately

In Motion Control, upload the action video and character image in their corresponding fields.

Check which reference guides the person’s orientation.

That orientation is not the file’s vertical or horizontal aspect ratio.

When face matching is available, it helps guide the face; clothing and accessories still need their own review.

Generate a short attempt within your limit.

Watch the entire gesture and check the hands, face, clothing, and background.

If only part of the action transferred, note the difference.

A similar expression does not prove that the full movement was preserved.

Otávio uses a tripod to keep the camera steady while recording.

3Check the decision against its use

Compatible framing leaves room for the gesture, not just the same body area.

Otávio uses the worksheet to check: review: gesture, face, hands, and clothing.

1The profile fixes what the audience recognizes

Define a working name, role, appearance, voice, and way of speaking.

For Nina: a virtual workshop presenter with short curly hair and a gray streak, a rust-colored jacket, and a calm voice.

Separate these traits from the context, such as the workbench, topic, and gesture.

The character’s story should serve the content without inventing real credentials.

A so-called master prompt is just reusable base text.

It helps you remember decisions, but it does not guarantee the same face.

Also save an approved reference image.

Compare the eyes, nose, hair, and clothing when reviewing variations.

Bia changes the scene topic while keeping Nina’s visual reference.

2Reuse the reference in your chosen feature

OpenArt documentation recommends starting from a main reference, preparing angles, and saving the character for reuse.

This keeps you from relying on a new description for every attempt.

Check the controls and select the saved character when the mode allows it.

A sheet of angles may also contain differences.

Review each view before adding it to the library.

If the hair or face changes, remake that image instead of teaching the next step the wrong variation.

For a real person, keep their own photos for comparison.

Otávio uses his own photos to check the digital version’s appearance.

3Check the decision against its use

An image from another angle goes into the library only after comparison with the main reference.

Otávio uses the worksheet to check: evidence: approved and reviewed reference.

1Menus make choices explicit

In Higgsfield AI Influencer Studio, open character creation and choose the attributes available in the Builder panel.

Start with a few visual choices and generate a proposal after checking the cost.

Save the version that best fits the character’s role.

For this exercise, imagine an alternative to Nina: a presenter in a blue shirt and green glasses.

It is a separate proposal, not a continuation of the previous identity.

Distinctive details can make a face recognizable, but they do not need to become a caricature or rely on stereotypes.

Bia records the alternative separately so it does not alter Nina’s reference.

2A saved character is not the same as a trained identity

Studio lets you save characters and send them to other features.

Soul ID is a separate feature that trains an identity from a set of photos.

The guide we checked asks for at least twenty images of the same identity.

Plan this step only after checking that the set is consistent.

For an invented character, you need to create and review the references.

Do not mix faces just to reach the required number.

The quality of the selection matters.

This course does not require training Soul ID to make your first video; the comparison helps you decide what to invest in next.

Otávio distinguishes saving an image from training an identity with multiple files.

3Check the decision against its use

A saved proposal, a set of references, and a trained identity are different states.

Otávio uses the worksheet to check: status: proposal saved or separate training.

1Four episodes, one identity

Plan an introduction, a tip, an explanation, and a closing.

Keep the reference, voice, and speaking style consistent.

Change only the topic and the needed gesture.

Each episode needs a short message that fits the rehearsed time.

Keep the text and settings with the final file.

At Oficina Clara, Nina introduces herself, shows a brush, explains a choice, and invites viewers to continue.

This organizes the series without inventing a personal testimonial.

If you use your digital version, check that the message is still something you want to say.

Bia keeps the text version that matches each approved audio file.

2Three tests show what you learned

Gather the talking avatar from HeyGen, the speech test in Kling, and the motion transfer using your own reference.

For each one, record the input, settings, cost, file, and review.

If a feature was unavailable, mark it unavailable or pending.

An alternative may serve the message, but does not prove the same operation.

Before sharing, watch on your phone and check the text, pronunciation, mouth, hands, and identity.

Clearly label a virtual character when applicable.

Keep approved files separate from tests.

The project is complete when the chosen deliverables exist and have been reviewed.

Otávio marks an episode ready only after watching the file on his phone.

3Check the decision against its use

Planning can be complete while videos are pending. Mark production complete only after watching the final files.

Otávio uses the sheet to check: delivery: reviewed files and visible pending work.