HeyGen Free Plan: What One Account Could and Could Not Show

Last updated: July 29, 2026

HeyGen is a browser-based platform that generates presenter-style video from a script, using stock or custom avatars. Paid web plans issue a monthly credit allocation and draw against it at published rates. The free tier is the route the vendor offers for trying that before paying.

FSR opened one free account and worked it across two sessions, two days apart.

The free account exposed the workflow and left the two questions that decide a purchase open.

Those two are what a minute of equivalent paid production would cost, and what a delivered video file would contain. Neither is a gap in the session. Both are properties of what a free account can reach.

Trial or plan

The search term most readers use says trial. The billing surface in the account said ‹Your Plan: Free›.

No captured surface showed an expiry notice, a countdown or a conversion prompt, and the upgrade control read as an upgrade rather than as a trial conversion. A one-day test cannot establish whether that state persists, and no expiry was waited out.

There is a second reason not to settle it here. HeyGen’s own materials describe the base tier as a plan while describing access to some premium features in trial terms. Whether a given capability sits inside the plan or inside a limited trial is a per-feature question, and this review does not answer it feature by feature.

Can free usage predict paid production cost

This is the question a buyer needs closed before paying, and the one a free account is least equipped to close.

Three observations bear on it. They are recorded separately because the account did not connect them.

The Free column also lists access to Avatar IV and to the prompt agent. HeyGen’s own pricing page publishes a per-minute credit rate for both of those workflows. The Free column publishes no credit figure. Those are three readings from two of the vendor’s own surfaces, and this review sets them next to each other without drawing a line between them.

Two things follow, and they are different in kind.

The first is arithmetic. A per-minute rate can be derived from the tooltip. Any such figure remains a calculation from one displayed allowance, not an independently verified rate, and it is not a second source corroborating the first.

The second is a gap. The paid tiers are described in credits. The free tier is described in its own unit. Whether a conversion between the two is published on any surface a buyer can reach is ‹the result of a search whose scope has to be recorded›, not an observation, and it does not publish as a bare statement of absence.

HeyGen’s July 2026 product update describes upfront estimates for actions that consume Premium Credits. The post does not clearly define how that applies to free plan access. This review reports only what the authenticated free composer displayed.

The second session tested the same question on a different surface. The prompt agent accepted a job, ran it for close to twenty-nine minutes and delivered it, and at no point in that sequence did the account display a cost. HeyGen’s own pricing page publishes a per-minute credit rate for that workflow.

If the two tiers are counted in different units and no conversion is reachable, time spent free produces interface familiarity rather than a cost estimate.

Three generations, two modes

FSR submitted three generations from this free account. Two produced completed assets. One returned an over-length error in the interface. Whether that request entered a render queue was not determined.

The first two came from the script composer on 28 July. The third came from the prompt agent on 30 July, and it is the one that changes the picture.

The prompt agent ran on a free account

On 30 July FSR opened the home surface, pressed the control that inserts a sample prompt, and submitted what the vendor supplied. The prompt asked for a thirty-second video on a named subject, with a stated visual style, avatar type and music direction. FSR wrote none of it.

The job completed. The interface reported the elapsed generation time as 1714 seconds, close to twenty-nine minutes. The finished asset ran thirty-seven seconds, across six scenes, and the account created two artifacts from the one prompt: the video, and a separate object the interface labels a video plan.

Those readings reconcile, which is worth saying in a review that spends a section on readings that do not. The timeline named six scenes, the export panel and the player both reported thirty-seven seconds, and playing the asset through showed six scene changes spanning about that length. Where these surfaces describe the same asset at the same grain, they agreed.

The download panel for that asset offered three tabs, for video, captions and audio. Both routes that would deliver the video file, direct download and save to cloud drive, carried the premium marker. FSR reports the markers as displayed and does not report a wall for this session, because the result of pressing them was not captured.

Two things are worth separating here rather than joining.

HeyGen’s published pricing lists the prompt agent workflow at a per-minute credit rate. This account has no published credit allocation. The account displayed no cost for the job, before submission or after completion. Those are three statements from three surfaces, and this review does not compute a figure from them.

The other is a ratio a buyer can use directly. Twenty-nine minutes of generation produced thirty-seven seconds of output, and the output stayed inside the account. On a plan counted in whole videos rather than minutes, that is what one unit of the allowance buys.

The composer accepted the over-length script for submission. The interface then displayed the message above. FSR did not inspect network or server logs and does not attribute the outcome to a particular system layer.

Whether the second attempt consumed any allowance is not determined. No counter relevant to the free video allowance was identified on the surfaces inspected, so no debit could be read either way. Not recorded is not the same as not consumed.

The length constraint applied after the script was written and the settings fixed, not before. On a metered plan that ordering decides whether a mistake costs a minute or an allowance, and the account gave no way to tell which one happened.

Can the asset leave the account

No video file was obtained through the controls tested on this account. That does not establish that every HeyGen free account, workflow or entry point behaves the same way.

One control carrying no premium marker was gated when used. The marker did not predict the gate here. One counterexample establishes that for this account on this date. It does not establish that the markers are unreliable in general, and this review does not say so.

The one file that came back is worth reading carefully rather than treating as a win. The interface reported the asset at ‹about seven seconds›. The delivered subtitle file ended its final cue at ‹three seconds›, with the whole utterance inside a single cue. These are two different measurements of different things, and this review does not treat them as contradictory without a shared asset identifier and a stated measurement definition. What it does mean is that the subtitle file is not usable as delivered.

The second completed asset, from 30 July, was not put through the same six controls. Its download panel was opened and photographed. It offered video, captions and audio tabs, and both routes to the video file carried the premium marker. FSR did not capture the result of pressing either, so this session adds a marker observation and not a second gate observation.

FSR could inspect hosted playback but not a downloaded video file. Delivered resolution, delivered duration, framerate, watermark burn-in and audio encoding stay outside this test. A reader deciding on the strength of a preview is deciding on a rendition the platform controls end to end. That held on both assets, two days apart, on two creation modes.

One thing the preview did allow. Watched in the platform’s own player, on the prompt-agent asset, lip movement tracked the delivered speech without drift visible to this observer. That is one asset, one avatar, one voice and a script FSR did not write, judged by eye rather than measured, in the vendor’s player rather than in a file. It is not a comparison with any other product.

It is worth stating for a structural reason rather than as praise. The dimension a free account can assess is the one a buyer could mostly have guessed at. The dimension it cannot assess is the one that decides the invoice.

Values that differed across surfaces

Named surfaces in one account displayed different values on the same day. The evidence did not identify an authoritative one.

A buyer planning against a maximum length has to choose a number the account does not settle. The verification cost falls on the buyer, across surfaces they would have to know to check. Recording the disagreement is the finding. Asserting a reason for it would be a second claim requiring evidence that does not exist here.

Free account against a structured paid test

A buyer whose open questions sit in the lower four rows may require a structured paid test rather than more free usage.

Terms and output rights

HeyGen’s terms of service address free plan output in one section and paid plan output in another, and the two are drafted differently. The free plan section describes the licence granted over free output in restrictive terms and lists prohibited uses. The section covering paid plans is drafted the other way. The free plan section also reserves the ability to impose or change limits, including export capability.

This section reports what the document states as of the access date below. It reaches no legal conclusion, characterises no conduct as permitted or prohibited for any reader, and offers no advice. A reader whose decision depends on output rights should read the current terms and take their own advice.

Tested and untested

None of the above is a claim that HeyGen publishes no answer. No reproducible documentation search was run for these items. Untested and documented-as-absent are different statuses.

FAQ

Methodology

This is a Tier B, hands-on review. FSR created one HeyGen free account and worked it across two sessions, submitting three generations across two creation modes and trying every file, cloud and sharing control the interface presented.

The interface changed between the two sessions. On 28 July the home surface presented a four-step setup card. On 30 July the same account presented a three-step card with different step labels, and one step present on the first date was absent on the second. FSR did not establish whether this was a product change, a state-dependent display, or something specific to this account or region. Anything this review says about the setup card describes 28 July.

Vendor documentation is recorded as an official claim, not as verification of runtime behaviour. Prices, plans, quotas, feature availability and terms are volatile, and were rechecked on the publication date.

Verdict

The free account exposed the workflow. It did not close what equivalent paid production would cost, and it did not close what a delivered video file would contain.

Those two gaps are structural rather than incidental. No paid plan was entered, so no debit was observable. No video file came back through the controls tried, so no file specification was inspectable.

Against that, a free account does not have to reproduce every paid feature to be worth an hour. A buyer who only needs to see the editor and hosted playback can get what they came for.

This review did not assess overall value, billing accuracy, output quality against competitors, or product intent. A buyer whose open questions are cost at volume or delivered file contents may require a structured paid test.