Most reviews of Synthesia argue about whether the avatars look real enough. That question got boring around 2024, and it was never the one that decides whether the tool works for you. Two things do: how the minutes are metered, and the fact that Synthesia reviews your videos before they exist.

That second point is unusual enough to lead with. This Synthesia review is built around it, because it’s the difference between a tool that’s perfect for corporate training and one that will quietly reject your third upload with no appeal.

What you’re actually buying

Synthesia turns a script into a video of a presenter saying it. You pick an avatar, type or paste text, choose a language, and the platform renders a talking-head video with slides, captions and branding around it. The company reports 240+ avatars and 160+ languages.

It does not edit footage you shot. It does not generate cinematic scenes from a prompt. It has one job — a person on camera delivering words — and it does that job at a quality most competitors haven’t matched. If you need a timeline editor, that’s a different category entirely, and Descript vs CapCut covers it.

The meter: minutes, per seat, no rollover

Synthesia bills in video minutes per month, not per video and not per user-hour. Ten minutes means ten one-minute clips, or two five-minute explainers, or one ten-minute training module. Unused minutes do not roll over.

The published tiers, as commonly reported in 2026:

Plan Reported price Minutes What it unlocks
Free / Basic $0 3–10 per month (sources disagree) Watermarked, limited avatars, testing only
Starter ~$29 monthly, ~$18–22 annual ~10 per month Watermark removed, download, 125+ avatars, up to 3 personal avatars
Creator ~$89 monthly, ~$53–67 annual ~30 per month 180+ avatars, up to 5 personal avatars, API access, interactive video
Enterprise Custom Unlimited SSO/SAML, SCORM export, brand kits, workspaces, live collaboration

An honesty note about that table. Trackers disagree substantially on Synthesia’s annual rates — Starter is variously reported at $18, $22 and $29, Creator at $53, $64, $67 and $89, and the free tier at both 3 and 10 minutes. At least one reviewer noted that Synthesia’s own plan cards and the FAQ on the same pricing page showed different annual totals on the same day. Treat every price you read, including the one above, as a starting point and confirm the number at checkout before you budget. That disagreement is a fact about the market, not a gap in any one review.

Now the arithmetic nobody prints. Starter at $29 for 10 minutes is $2.90 per finished minute. Creator at $89 for 30 is about $2.97. The per-minute cost barely improves as you move up — you’re buying headroom, not a discount.

The thing every review misses: minutes are per seat

This is the single most expensive misunderstanding available.

Starter and Creator include one editor seat. Guests can review and comment; anyone who actually creates or edits a video needs their own paid subscription. And the minutes do not pool across seats — each seat gets its own allowance, so one heavy user hits the cap while a light user’s minutes expire unused, with no easy way to move the surplus.

So a five-person learning-and-development team on Creator is not $89 a month. It’s roughly five Creator subscriptions — around $445 at the monthly rate. Reported overage minutes run about $2–5 each on top, and custom studio avatars have been reported at around $1,000 per avatar per year.

The rule of thumb: multiply the headline price by the number of people who will actually click “generate,” then ask whether Enterprise’s unlimited minutes are cheaper than four or five stacked self-serve plans. Frequently they are, which is presumably the point. Marketplace data compiled from a small sample of verified enterprise purchases puts median annual spend in the region of $30,000 — useful mainly as a signal that Enterprise is a real budget line, not an upsell you can shrug off.

This is the same billing-model trap that decides whether AI actually saves money, laid out more generally in cutting business costs with AI.

The moderation queue

Here’s what makes Synthesia genuinely different from every other tool covered on this site: it moderates content at the point of creation. Not after a complaint. Before the video renders.

Synthesia’s own documentation describes technical filters plus human oversight, with videos examined case by case on both topic and use case. Videos get rejected. You can edit and resubmit, but the company states plainly that resubmission doesn’t guarantee approval. Accidental breaches typically draw a rejection with an explanation; repeated or serious violations can restrict features or the account (Synthesia’s content moderation FAQ).

Three specifics that matter commercially:

  • Stock avatars are the most restricted. They’re real actors who licensed their likeness, and Synthesia limits what topics their faces can be attached to. Personal and selfie avatars — where the face is yours — get more latitude.
  • News-style content is gated to enterprise accounts. After researchers documented Synthesia avatars appearing in state-aligned disinformation videos, the company broadened its ban beyond misinformation to polarising content and restricted news-format output to corporate customers.
  • Political content is similarly restricted to corporate accounts, framed by the company as election-integrity policy.

Whether this is a feature or a problem depends entirely on who you are. For a regulated enterprise, moderation at creation is exactly the control that makes avatar video approvable by legal. For an independent creator or a marketing agency with unusual clients, it’s a dependency on someone else’s judgement sitting between you and publishing — on a deadline.

The practical read from Synthesia’s own guidance: the only route to predictable approval is a studio or personal avatar. If your work will run at volume, don’t build it on stock faces.

Creating a personal avatar requires a sample plus a consent video containing a spoken statement and a passcode. Synthesia extracts and matches data points — face geometry, voice properties — between the two to confirm the same person supplied both. Photo avatars are rejected outright if the identity doesn’t match or consent isn’t unambiguous, and the source photo must be a real person: no AI-generated images, no illustrations, no celebrities, no deceased people (photo avatar requirements).

Synthesia’s privacy policy is explicit that this is biometric processing, and cites both a GDPR special-category derogation and Illinois biometric-privacy law as reasons it requires explicit consent (privacy policy).

Two consequences worth thinking about before you put your team’s faces into it.

First, the good half: the person owns the revocation. An avatar owner can request deletion of their likeness at any time, and Synthesia states deletion is permanent with confirmation, typically within 90 days (AI governance page). If your CEO’s avatar fronts your onboarding library and your CEO leaves on bad terms, that library has a dependency you didn’t budget for.

Second, the part that needs a policy: an employee’s face and voice are biometric personal data, and “we’re doing an AI video” is not a substitute for a real, documented, revocable consent conversation. Synthesia’s process handles the platform’s obligation. It does not handle yours as an employer.

This is a different legal problem from the one in the ElevenLabs review, where the fight is over cloning voices you may not have the right to use. Here the platform enforces consent for you — the exposure is what happens inside your own organisation, adjacent to the issues in AI for HR.

Where it’s genuinely worth the money

Use case Verdict
Internal training and onboarding at scale Strong buy. SCORM export, updateable scripts, and moderation that legal will like
Multilingual versions of the same content Strong. The clearest advantage — one script, many languages, no reshoot
Product explainers and help-centre video Good. Content that changes often and never needed a real presenter
Sales personalisation at volume Conditional. Works technically; the minute meter and recipient reaction are both risks
Social content for a creator brand Weak. Wrong economics, wrong format, and moderation in the loop
Anything news-shaped or political Enterprise only, by policy

The multilingual case deserves the emphasis. Re-recording a training module in eight languages is a genuine budget line that Synthesia removes. That’s the same economic logic that makes dubbing the highest-return AI use for content creators on YouTube and TikTok — you’re multiplying the value of work already done rather than manufacturing more.

The weak case deserves the emphasis too. At roughly $3 a finished minute, with a monthly cap, an avatar-led social channel is expensive per upload and structurally disadvantaged on the platforms. Synthesia has never really been a creator tool, whatever the marketing suggests. If that’s what you’re after, making AI videos for social media without editing skills is the more realistic starting point.

Honest limitations

The avatars are good, not invisible. They handle straight delivery well and struggle with emotional range, comic timing and anything requiring a physical demonstration. Audiences increasingly clock them.

Rendering is a queue, not an instant. Fine for a planned module, awkward for a same-day fix.

The script is the product. Synthesia removes production friction and adds none of the thinking. A bad module rendered by a photogenic avatar is still a bad module.

Minutes expire monthly. If your production is lumpy — nothing for six weeks, then a whole course — the meter fits badly, and annual billing doesn’t create a shared annual pool at the self-serve tiers.

Enterprise is where the useful features live. SSO, SCORM, workspaces, brand kits and collaboration are gated. If you need any of those, the self-serve tiers are a trial, not a plan.

The verdict

Synthesia is not really an AI video generator competing on realism. It’s a corporate communications system that sells governance — consent verification, moderation at creation, avatar ownership controls, audit-friendly deletion — with a video renderer attached. Priced and judged that way, it’s coherent and often worth it.

Judged as a creator tool, it’s expensive per minute, capped monthly, single-seat by default, and it reserves the right to reject your work. That’s not a flaw. It’s the product working as designed, for someone who isn’t you.

Buy it if you produce training, onboarding or product video in multiple languages and someone in your organisation has to sign off on what goes out. Skip it if you’re making content for an audience that came for a person. For where it fits in a wider stack, the best AI tools for small business owners has the map, and AI for elearning and course creators covers the surrounding workflow.

Frequently asked questions

How much does Synthesia actually cost per video?

At roughly $29 a month for about 10 minutes, Starter works out near $2.90 per finished minute, and Creator lands at a similar per-minute rate. The rate barely improves as you move up tiers, so upgrading buys headroom rather than a discount. Reported annual rates vary widely between sources — confirm the total at checkout.

Do Synthesia minutes roll over or get shared across a team?

Neither. Unused minutes expire at the end of the billing cycle, and on Starter and Creator each editor seat has its own separate allowance that cannot be pooled. A team of five needs five subscriptions or an Enterprise plan.

Can Synthesia reject my video?

Yes. Synthesia moderates at the point of creation, using automated filters plus human review, and examines videos case by case on both topic and use case. You can edit and resubmit, but approval isn’t guaranteed. News-style and political content is restricted to corporate accounts.

What does creating a personal avatar involve?

A sample and a separate consent video containing a spoken consent statement and a passcode. Synthesia matches biometric data points between the two to confirm the same person provided both. Submissions are rejected if identity doesn’t match, if consent is ambiguous, or if the source image is AI-generated, a celebrity, or a deceased person.

Who owns and controls an avatar of an employee?

The person whose likeness it is retains control, including the right to request deletion at any time, with Synthesia stating permanent deletion typically within 90 days. That means video libraries fronted by a departing employee’s avatar carry a dependency worth planning for, and employers need their own documented consent process regardless of the platform’s.

Is Synthesia worth it for a solo creator or small channel?

Usually not. The per-minute cost, the monthly cap, the single editor seat and moderation in the publishing path all point toward organisations rather than individuals. It’s strongest for internal training, onboarding and multilingual product content where approval workflow is a requirement rather than an obstacle.