How to Make a Talking Avatar Video

A talking avatar video is a video of a portrait speaking an audio track. You do not need a camera, an actor, or a timeline. You need one clear portrait, one clean voice track, and a short script that earns the viewer's attention.

This guide walks through the full workflow in GoodAvatar, from choosing the portrait to downloading the final MP4. It also covers the mistakes that most often make a clip look "off" and the cost of a typical 30-second video.

What you need

• One portrait: JPG or PNG, clear face, even lighting.
• One voice track: MP3 or WAV, clean audio, within the 120-second single-track limit.
• A short script, if you are generating the voiceover with text to speech.
• Credits: 5 credits per second; new accounts get 100 credits.

If you do not have a portrait or a voice track, the create flow can generate both. The portrait helper and text-to-speech helper are optional; video generation is the step that reserves video credits.

Step 1: Choose a portrait that reads well

The portrait is the anchor of the whole video. A sharp, front-facing image with even lighting usually beats a heavily retouched studio photo.

What to look for:

• Face large enough to read expression
• Eyes visible, no strong shadow across the face
• Front-facing or a slight angle
• Natural skin texture, no heavy beauty filter
• Simple background that does not pull attention

What to avoid:

• Extreme wide-angle distortion
• Sunglasses or a hand covering the face
• Very small face in a large frame
• Heavy HDR or artificial sharpening

If the face is small, generate the portrait at a closer crop. Only AI-generated portraits are accepted — real photographs are rejected before generation.

Step 2: Prepare the audio

The voice track drives the timing. A clean 20-second track usually produces a better result than a noisy 90-second track.

• Use MP3 or WAV.
• Remove loud background music and clipping.
• Keep the first test at 15–30 seconds.
• Stay within the current 120-second single-track limit.

You can upload your own recording or generate a voiceover from a script with the built-in text-to-speech helper. If you use your own voice, keep the same microphone and room across a series of clips so the presenter feels consistent.

Step 3: Write for the ear

If you are generating the voiceover, write the script for listening, not for reading. Short sentences. One idea per sentence. Avoid visual references such as "as you can see here" unless the video actually shows it.

A simple 30-second structure:

1. Hook (3–5s): the problem or the reason to keep watching.
2. Context (8–12s): what the viewer needs to know.
3. Proof (8–12s): a concrete detail, number, or example.
4. Next step (3–5s): what to do now.

For a 30-second clip, 60–90 words is usually enough. Read it aloud and cut anything that does not earn its place.

Step 4: Review the estimate and generate

The create page shows the estimated duration and the credits that will be reserved. A 30-second video uses 150 credits. The task then runs in the background; you can close the browser and return from the task list or the email notification.

Generation is roughly 20–25 times the video length. A 30-second clip usually takes about 10–15 minutes.

Step 5: Review the output

Check four things:

• Lip sync: do the mouth shapes match the voice?
• Face stability: does the face drift or distort?
• Framing: is the head cropped in a natural place?
• Audio: is the voice clear and at a consistent volume?

If the result is close but not right, change one input at a time. A different portrait often fixes face stability; a cleaner audio track often fixes sync.

What it costs

GoodAvatar uses 5 credits per second:

The Starter pack is $8.99 for 800 credits (160 seconds). Credits are valid for 180 days. If a generation fails, the reserved credits are returned automatically; there are no exceptions. There is no subscription, the output has no GoodAvatar watermark, and you can use it commercially.

Common mistakes

Using a portrait where the face is too small. The model has less information to work with. Crop closer.

Using audio with loud music. The voice is the signal; the music is noise. Use a clean track.

Writing a 90-second script for a 30-second slot. Cut the script before generating, not after.

Changing five things at once. If the first result is off, change one input per run so you can learn what matters.

Expecting a full video editor. GoodAvatar creates the talking-head clip. Use your normal editor for B-roll, captions, and scene changes.

Create a talking avatar for free

You can test the workflow without buying a pack. A new account includes 100 credits, enough for about 20 seconds of video. No paid plan is required.

1. Upload a clear portrait, or generate one from a prompt.
2. Upload a 15–20 second voice track, or generate the voiceover from a script.
3. Review the estimate and submit the task.
4. Download the MP4 when it finishes.

A 20-second test costs 100 credits, so the free balance covers one clean test. Keep the script short and the audio clean; the goal is to judge the face, the sync, and the framing before spending money.

If the test works, the Starter pack is $8.99 for 800 credits (160 seconds), enough for five 30-second clips. Credits are valid for 180 days, failed generations return all reserved credits, and the output has no GoodAvatar watermark. You can use the clip commercially, subject to the rights you have in the portrait and audio.

For the plain-language definition of the format, read what is a talking head video. For the full feature page, see the AI avatar video generator.

Where to use the clip

• Product explainers and launch notes
• Course intros and lesson summaries
• Sales follow-ups
• Help-center walkthroughs
• Social clips built from a repeatable script

For a deeper workflow page, see the AI avatar video generator. For the photo-specific guide, see photo to talking video. For pricing, see pricing.

Start with a 30-second test

Use a clear portrait, a clean 20–30 second track, and a short script. Create your first video with the 100 starter credits. If the result fits, keep the same portrait and voice and update the script for the next clip.