How to Test a Talking Avatar With HeyGen Free—Before You Pay

Free-plan documentation checked October 8, 2026. The example video and screenshots are from August 26, 2026.

Check what your Free account actually includes

HeyGen's pricing page advertises three Free videos per month, videos up to one minute, and limited access to premium features such as Avatar IV. You can start without a credit card. Its plan explanation adds an important qualification: the Free quota is one to three videos depending on region, and the output it lists is a 720p sharing link. Free uses a video quota rather than the paid plans' credit balance.

The Photo Avatar guide lists a narrower Free Avatar IV allowance: ten seconds per video and thirty seconds per month. In our August test, the signed-in account displayed three watermarked videos up to one minute and completed a 14.466-second Avatar IV clip. That historical result does not establish what your account will permit today.

After signing in, open the plan or usage display and check the following before uploading a photo or generating a video:

• The account is on Free, and the allowance has videos remaining.
• The selected avatar engine is included. A visible model name or premium badge does not mean you can generate with it for free.
• The duration fits both the overall video limit and any limit for that model or Photo Avatar route.
• The watermark and available sharing or downloading options suit the test you intend to run.

If a generation button leads to a paid plan, purchased credits, or a trial checkout, return to the editor and choose an included option. If none is available, finish the audio and layout preparation and wait for the allowance shown by your account. A trial requiring payment details is a separate purchase decision.

If you have only one generation left, finish the pronunciation and timing checks below before using it. If you have several, you can use a later attempt to compare one change, such as another voice or portrait. Creating several scenes does not establish that you have several free generations available; read the account's remaining allowance.

Choose the avatar that answers your question

You need HeyGen, a short script, and a browser for the basic test. Image generation, voice cloning, local transcription, and command-line tools are optional. Choose your avatar according to what you want to learn:

Your question	Useful starting point	What the result can tell you
Does this script work when spoken by an avatar?	A stock avatar available on Free	Speech clarity, pacing, layout, and the basic generation process
Will this particular portrait animate acceptably?	Your authorized photo or the supplied fictional portrait	How that face and crop behave with the selected engine
Can I use a fictional brand presenter?	A synthetic Photo Avatar with a readable face	Whether the character's mouth, expression, and framing suit the message
Will a digital version of me sound and move like me?	A separate Digital Twin and voice evaluation	A stock voice plus one still photo cannot answer this identity question

A stock avatar is a useful first choice when the script is your main concern. If you later change to your own face, review that output separately: a successful stock-avatar clip tells you little about the stability of a different portrait. Keep the script and voice unchanged during that comparison so you can see what the avatar change affected.

Prepare a clear photo

HeyGen recommends visible eyes and lips, recognizable facial proportions, and a character that occupies enough of the image to detect clearly. A tiny face in a wide photograph, a mask over the mouth, or an exaggerated snout makes a poor first test. Use a clear portrait with room around the head and shoulders; inspect the mouth and face outline before uploading.

For your own photo, check the background for documents, badges, screens, and other people. For someone else's face, establish permission to animate the image and use it in the intended video. A stock voice avoids needing a voice recording for this exercise. Keep the presenter visibly or audibly identified as synthetic when viewers could mistake it for a real person's recorded performance.

You can use the fictional adult portrait in the example files. It was created with OpenAI ImageGen. You do not need an image-generation subscription or a new portrait to follow the guide.

Write a script that exposes useful problems

Test a sentence you might actually use. A greeting alone will not tell you whether the voice can say your product code, a person's name, or a price correctly. Choose the hardest meaningful line from your planned video and include enough surrounding words to judge its delivery.

Before editing, make a small table of facts that must survive. The supplied example is entirely fictional:

Fact to preserve	Spoken form in the tested script	Listen for
Maya Chen	Maya Chen	Both names, without a dropped surname
Atlas B72	Atlas B seventy-two	The letter B and the correct number
$48, not $84	Forty-eight dollars, not eighty-four	The amount and the word "not"
Thursday at 3:30 p.m.	Thursday at three thirty p.m.	The weekday, time, and p.m.
A synthetic presenter	This presenter is synthetic.	The disclosure remains in the message

The first draft in the recorded test changed the price to "about fifty dollars," shortened Maya's name, dropped B72 and the review time, and called the avatar ready to share. Those changes made it unusable even though the sentence read smoothly. The corrected script was:

Hello, I am Maya Chen. This presenter is synthetic. Atlas B seventy-two
costs forty-eight dollars, not eighty-four. Review Thursday at three thirty p.m.

The full example contains 23 words and took 14.466 seconds with the tested voice. If your Photo Avatar allowance is ten seconds, try a shorter line such as "Atlas B seventy-two costs forty-eight dollars, not eighty-four." Preview its duration with your chosen voice and keep the synthetic disclosure on screen. This suggested shorter line has not been rendered for this article.

Keep editing the wording without losing the facts

Speak the script aloud. If a sentence feels crowded, split it at a meaningful boundary or remove a nonessential claim. Keep the exact name, amount, date, condition, and negation. For a real appointment, include the calendar date and time zone when "Thursday at three thirty" would leave the audience uncertain.

If you ask AI to rewrite the line, give it the facts separately and compare the result yourself. A useful request is:

Rewrite this for spoken delivery. Preserve every name, product code,
number, date, time, and negation exactly in meaning. Do not add claims.
Return the revised script and list any wording whose meaning is uncertain.

Facts that must survive: [paste your fact table]
Script: [paste your draft]

The example fact list shows what was checked in Maya's clip. A note on your phone or a small spreadsheet works too. Replace the example entries with the actual names, numbers, and instructions in your video.

Build one scene and preview the voice

AI Studio gives you a direct route to a scene with an avatar, script, and voice. Follow HeyGen's Studio walkthrough if you need help locating the controls. For this first test, use a blank scene so a template's extra elements do not obscure the face or your text.

1. Open AI Studio and create a scene. Choose portrait orientation for a vertical destination or landscape for a horizontal presentation.
2. Use the Avatar control in the right panel to select a stock avatar included in your account. For a photo, go to Avatars, choose New Avatar, upload the photo, name it, and wait for validation before selecting it in the scene.
3. Check that the scene contains an avatar. Adding a photograph through Assets places an image on the canvas; it does not create a speaking Photo Avatar.
4. Choose an engine available within your Free allowance. Keep the engine name in your notes, especially if you intend to compare results later.
5. Paste your reviewed script and select an available stock voice in the Voice section.
6. Play the script audio. Correct pronunciation, pace, and duration before submitting the video for generation.

You may also see Single Scene or Quick Create from an avatar. HeyGen's Single Scene guide distinguishes Presenter Mode, which uses a script or audio, from Cinematic Mode, which emphasizes prompted action and camera movement. Use the presenter route for a talking-head test, then check the actual engine and cost shown. Current documentation emphasizes Avatar V and credit costs, so the old Single Scene screenshot below should not be treated as a guarantee of free access.

Audio preview does not preview the moving face

HeyGen's voice guidance says Audio Preview is free, although excessive use over extended periods can trigger temporary safeguards. It lets you check narration and timing. Lip sync, facial expressions, and avatar motion require a rendered scene, so a still face while listening is expected.

Changing the script, voice, or voice settings causes the next preview to generate new audio. Listen after the last edit. A voice sample that sounds good on its own may still pronounce your name or code badly when reading the actual sentence.

If an old tutorial sends you to a homepage Voice tab, HeyGen has moved voice access into places including AI Studio and the avatar page. Voices are assigned at avatar level by default, so check every scene after changing one. To use different voices per scene, enable that option from the Studio's top-left menu.

If the voice sounds flat, rushed, or wrong for the audience

Compare available stock voices using the same actual sentence. Listen for the accent your audience expects and whether the delivery suits a routine update, lesson, or promotion. Remove stiff wording before trying to make the voice more expressive. A sentence crowded with product claims often needs editing even when another voice sounds warmer.

Studio provides volume and speed controls. Adjust speed gently and replay difficult names and numbers after changing it. Some voice engines have extra style controls; use only settings your account makes available. Buying more video allowance is not a substitute for finding a voice that reads your script well.

Fix names, numbers, pauses, and language

In HeyGen's script editor, double-click a mispronounced word and choose Pronunciation to enter a phonetic spelling. That correction applies to the particular occurrence in that scene; check repeated uses separately. Write numbers in the intended spoken form when the preview reads them incorrectly. "Three thirty p.m." and "forty-eight dollars" give you more control than ambiguous numeric formatting.

Place the cursor at a sentence boundary and use Pause if the available control helps the message breathe. The documented adjustment is in half-second intervals. Listen again after adding it, since pauses count toward the time you are trying to fit.

Keep a scene in one language where possible. HeyGen warns that mixing languages in one script can produce unpredictable voice output. For a second language, use a separate scene or project with an appropriate voice and have a fluent listener check the result. A successful English test does not establish the quality of your Chinese or Arabic version.

If the script exceeds your time limit

Check the duration after choosing the voice and adding pauses. The tested 23-word script took over fourteen seconds, so the same word count cannot be assumed to fit a ten-second Photo Avatar limit. Choose one difficult sentence for the first render and save the rest of the message for a later project.

Cut a greeting, repeated benefit, or background explanation before cutting the condition that makes a price or instruction accurate. If the remaining line still takes too long at a comfortable pace, use a shorter test or plan for a different allowance. Increasing speed until the name or code becomes hard to understand defeats the purpose of the test.

Generate your first video

Before submitting, confirm the selected avatar, engine, voice, script, duration, and orientation. Keep a copy of the script and note the displayed allowance. If your account provides only one video, use it on the sentence and portrait that matter most to your decision.

Generate once, then open the completed project. The August account started generation immediately when the button was clicked; it did not show a final cost dialog. Do not rely on receiving another confirmation screen. If the project is still processing, check its status before submitting a duplicate.

Use a simple comparison when deciding whether to try again. Keep the script fixed when testing another voice; keep the voice and wording fixed when testing another portrait. Changing all the inputs together may produce a nicer clip while leaving you unsure which change helped.

Leave elaborate motion prompts and background music out of a first speech-and-face test. They add things to judge before you know whether the avatar can deliver the difficult sentence. Once that works, test the extra elements against the same acceptance points. Keep enough room in the frame for captions rather than filling every edge with the presenter.

Review speech and motion separately

Play the completed video at normal speed once without reading the script. If you miss a name, code, or instruction, note the phrase and its time in the clip. Then play it again against the written facts. Reading a caption while listening can make an unclear word seem more intelligible than it is.

Check	How to inspect it	Reason to revise or reject
Meaning	Compare speech with every row of your fact table	A changed price, missing "not," wrong code, or omitted instruction
Voice delivery	Listen without captions, then replay the difficult phrase	The target audience cannot understand it comfortably at normal speed
Lip sync	Watch full-speed playback with sound after the video has rendered	The mouth repeatedly leads or trails the audio
Face and movement	Watch continuously, then pause where something looks odd	Persistent jaw distortion, unstable teeth, shifting face shape, or distracting gestures
Framing	Inspect the actual orientation and intended viewing size	Clipped head or shoulders, unreadable text, or captions over the mouth
Delivery	Check the watermark and official sharing or download controls	The available output cannot meet the project's destination requirements

For lip-sync concerns, compare another completed video in the same player and try the affected clip again on a stable connection. This helps distinguish a playback issue from a defect you repeatedly see in the generated result. If you legitimately have a downloaded file, check that version too. Do not diagnose the finished animation from an editor's static preview.

For a face defect, inspect the original portrait at the same area. If the mouth was obscured or the face too small, a clearer input is a reasonable next test. If the source is clear and the generated jaw repeatedly warps, record the problem and try one controlled change only if your allowance permits it. A different voice or more punctuation is not a proven repair for defective animation.

The downloadable checklist gives you a place to record results. Mark an unchecked item as unresolved. A clip with a correct price but an unclear product code still needs a decision about that code.

If the project fails, stalls, or has no sound

If generation reports an error, keep the error text and the project identifier privately. Check whether the account still shows an allowance before starting another attempt. If the cause is unclear, use HeyGen support with the project details; do not assume a failed job restores your quota. For a project still marked as processing, return to its status before submitting the same scene again.

If the completed player is silent, check its mute control and your device's sound, then compare another video. Return to the scene to confirm it contains the intended script and selected voice. If other videos play normally and this completed project remains silent, record that failure before deciding to regenerate. If the photo stays still, first confirm it was created and selected as an avatar rather than added as a canvas image.

What an automatic transcript can and cannot settle

In the recorded test, Whisper Small recovered every protected field. Whisper Base transcribed "Atlas B" as "At least B." The account owner listened to the in-account result at normal speed and confirmed Atlas B72 was clear, the audio was intelligible, and there was no obvious lip-sync lead or lag.

That disagreement was a reason to listen closely to the code. If you use transcription, compare the questionable word with what a listener hears in the video. A correct transcript says nothing about lip sync or facial movement. You can complete the review by listening and watching, without installing a transcription model.

Check sharing and downloading before committing to a project

HeyGen's current Free plan explanation lists sharing links. Its download guide describes downloading as a paid-user feature: open Projects, use the video's three-dot menu, and choose Download. In the tested Free account, that action opened a Creator upgrade prompt. Successful playback in HeyGen does not guarantee a free MP4.

Inspect the controls your own account offers. A HeyGen-hosted link and a local video file serve different needs. A link may suit a review when the viewer can use it and its access settings are acceptable. A slide deck, an upload to another platform, or a client's file handoff may require the actual MP4. Verify that requirement before making several scenes.

If Download asks for an upgrade, choose whether the paid export meets your needs or use a different production method. The tutorial's review kit contains images, scripts, and records, not an MP4 or a substitute export. No public share link was created for the historical test.

Check captions in the final output

HeyGen's caption guide describes stylized captions in AI Studio. Enable Captions from the right-side menu when available, edit the text, and position it away from the mouth and important on-screen information. Inspect the rendered video for the spelling of codes, names, and numbers. A pronunciation spelling used for speech may need different wording in the visible caption.

The guide distinguishes these rendered captions from closed captions, which it does not currently offer for Avatar Videos. If your destination needs a separate subtitle file, check the actual export options and destination requirements before choosing the workflow. For a vertical clip, review the caption placement at phone size as well as in the editor.

Look at the video at the size your audience will see. A tiny presenter beside a large slide tells you less about the face than a close-up. A tall portrait also needs a new framing check if you later turn it into a horizontal video. The example here was 720 × 1280; a higher export resolution cannot recover detail that was absent from the original image.

Adapt the test to your actual use

A product introduction with prices and codes

Use the product's real name, its hardest code, and one sentence with the price or condition viewers must understand. If the code can be read two ways, settle the intended pronunciation before generating. "B seven two" may distinguish individual characters better than "B seventy-two" for your audience; preview both with the same voice. The historical B72 clip tested only the latter form.

Add any condition that changes the meaning of the price, such as whether it is a monthly charge or includes a required fee. Put the same facts on screen and check them against the speech. A fluent presenter cannot establish that a product claim is true. For a longer demonstration, use the product-video guide to plan the supporting shots.

For a subscription product, try this structure with your verified facts: "The [plan name] costs [amount] per [billing period]. [Feature or limit] is included. [Important condition]." Listen for the billing period and condition as carefully as the price. Use real product footage when viewers need to see how the product works; a presenter reading a claim cannot demonstrate the feature.

Your own face in an internal update

Decide whether you want an animated photo with a stock voice or a representation that sounds like you. Those are different tests. For the photo route, compare the rendered expression and crop with what you are comfortable showing colleagues. Ask whether the voice's pace suits a routine update and whether the synthetic presentation is clear to the audience.

If the likeness is acceptable but the voice feels unsuitable, change the voice while retaining the script and photo. If matching your identity is essential, assess the Digital Twin and voice setup separately before investing in a longer video. For an update about a meeting, include the actual date and time zone so colleagues cannot mistake the timing in a forwarded clip.

A fictional character or illustrated mascot

First inspect whether the character has a visible mouth and readable eyes. Test speech before investing in extra gestures or camera movement. A highly stylized face may need a different drawing or engine; a good result with Maya's human-shaped portrait does not predict how a mascot will behave.

Choose a line with the character's intended delivery. A calm narrator and an excited promotional character may need different voices. Keep the first motion test simple enough to see whether the face stays consistent while speaking. Judge whether the character communicates the message comfortably, including any disclosure the audience needs.

A video in another language

Prepare the meaning and intended pronunciation of names in that language before selecting a voice. Have someone fluent check a representative line, especially numbers, negations, and product terms. When your planned video has two languages, have each version reviewed. A clear English voice does not tell you whether the second language sounds natural.

If your starting point is an existing recorded video, creating a new avatar scene and translating that recording are different workflows. Use the translation and dubbing guide for preserving the original video's meaning. This Free avatar test does not establish translation quality.

A course or a video longer than the Free limit

Select a sample from the part of the lesson that is hardest to deliver: a technical name, a change in emphasis, or an instruction learners must follow. Review that sample with the voice and avatar you plan to use. A pass gives you a useful reason to continue testing; it does not prove that a long lesson will remain comfortable to watch.

For the full lesson, decide where the presenter should be visible and where a screen recording, diagram, or demonstration better serves the learner. Estimate production needs from those avatar scenes, including revisions. If you need a longer script first, the fact-based video-script guide covers scripting and shot planning.

A client handoff or public campaign

Confirm the deliverables before generating: file or link, orientation, resolution, caption format, watermark tolerance, and who approves the exact script. Review a representative sample with the person who will accept the final work. Avoid a situation where everyone likes the face but the available output cannot be uploaded to the required destination.

Ask the client to check the sample at the intended viewing size and listen to its actual product names. Agree on the voice and pronunciation before building more scenes. Keep the approved script with the delivered file so a later revision can be compared with what was accepted. The example here was an editorial test, not a client delivery or campaign result.

What the example showed

On August 26, 2026, one Free account generated the fictional Maya portrait with the 23-word script, Avatar IV, and Ana - Cheerful. The watermarked result lasted 14.466 seconds at 720 × 1280. The account owner heard Atlas B72 clearly and found the speech intelligible with no obvious lip-sync lead or lag at normal speed. Fourteen sampled frames showed no sampled face or framing defect, although still images cannot assess the full motion.

Official Download opened a Creator upgrade prompt. The full test record and example ZIP preserve the inputs, screenshots, and measurements; they contain no MP4 or share link. The ZIP's README explains its optional technical checks. All screenshots are from the August test. The October 8 update checked documentation; it did not generate a new HeyGen video.

When another render or a paid plan makes sense

Keep a passing clip when it answers the question you set out to test. Use another allowance when you can name a problem and a change worth testing. "The price at this point is unclear; I will add a pause and preview again" is useful. If your concern is how the presenter feels to viewers, collect a specific comment about its voice, expression, or pacing before changing the inputs.

Result	Next decision
Speech, face, and framing work; you only need a test	Keep the result and save your settings. Stay on Free while that meets your needs.
The result works, but your project needs official download, watermark removal, or more duration	Compare the current paid features with those specific requirements.
The voice misreads a code in Audio Preview	Fix the pronunciation or spoken form before spending a render.
The rendered mouth repeatedly fails with a suitable portrait	Try a different available portrait or engine while keeping the script fixed. Stop if the repeated result is unsuitable.
The test is unfinished because the quota is exhausted	Wait for the account's stated renewal or assess a paid test budget. An unfinished test has not demonstrated suitability.
The delivery requires showing a real process or screen	Consider a recording or demonstration, with an avatar only where it helps explain the material.

Current Creator pricing is $29 per month with 600 credits, 1080p export, and watermark removal. Decide which specific feature your project needs before choosing a subscription. Higher resolution or removal of a watermark addresses delivery requirements; it does not show that a particular pronunciation or facial movement will improve.

To estimate the generation budget, HeyGen's credit guide lists Avatar IV Photo Look at sixteen credits per minute and Video Look at thirty-one. As planning arithmetic, generating a one-minute Photo Look scene twice would use thirty-two credits for those two generations before other charged features. Extra motion and other generation tools can add costs. Use the estimate shown in your account for the actual job; the arithmetic here is not a tested billing result.

If you need several videos, include likely revisions in that budget and confirm the required export, watermark, and caption options. Save a sample you are happy with before subscribing for a larger project. If you cannot yet get a clear voice and acceptable face, spend the next test on that problem before committing to a longer production.