HeyGen turns scripts, images, audio or slide decks into videos presented by an AI avatar, with voice cloning and one-click translation into other languages. It is used for marketing, training and localisation, and its policies require explicit consent before you create an avatar of any real person.
Used this tool? Rate it
Used this tool? Rate it
HeyGen generates videos in which an AI avatar delivers your script to camera. The homepage states the proposition directly: "HeyGen: Create Realistic AI Videos of Yourself in Minutes", with the supporting lines describing AI videos starring you and the idea of being everywhere without being everywhere. That second phrase captures the actual value: recording a presenter is slow and does not scale across languages or updates, whereas a digital twin does.
The workflow is straightforward. You supply a script — or an image, an audio file, or a slide deck — choose an avatar, and the system produces a video of that avatar speaking your content with synchronised lip movement. From there you can clone a voice, translate the finished video into other languages, or drive the whole process through an API.
The operating entity is disclosed clearly in the terms: HeyGen Technology, Inc., operating from 12130 Millennium Drive Suite 300, Los Angeles, with California law governing and a minimum age of 18. The company published its own funding announcement describing a $60M Series A led by Benchmark, valuing the company at over $500M, with Thrive Capital, BOND and SV Angel participating. The same post reports growth from $1M to $35M+ ARR in just over a year, more than 40,000 paying business customers and profitability by Q2 2023 — figures the company disclosed itself rather than independently audited numbers.
Vendor-reported scale claims on the homepage include 30 million users, customers across 196 countries and 85% of the Fortune 100. Treat these as marketing rather than verified metrics.
One clarification on this entry's target: the registered address is app.heygen.com, the product's application entry point. It resolves normally and belongs to the same product as the heygen.com marketing site; the policy and pricing documents cited throughout are published on the main domain.
This section comes early rather than buried in limitations, because it is the single most consequential thing to understand about this category of tool, and because it is easy to assume that anything the software can technically do is something you are permitted to do.
HeyGen's moderation policy is explicit: "You are required to obtain the explicit consent of the individual being represented" (the "Actor"), and unless explicit consent has been provided, creating avatars of other individuals is strictly prohibited. This is not buried boilerplate — it is the operative rule for the product's core feature.
The policy names public figures specifically. It prohibits content that would "Represent real individuals, including celebrities or public figures, without their explicit consent". The intuition that a public figure's image is fair game because it is widely available is wrong under these terms.
Other prohibited categories include illegal, defamatory, sexually explicit, pornographic, violent, threatening, abusive, inflammatory, harmful, hateful, cruel, insensitive or deceptive imagery, along with depictions appearing to represent individuals under 18 and copyrighted images used without the owner's consent.
The trust and safety page states the ownership and responsibility split plainly: you retain full ownership of avatars you create, but "you must ensure you have the legal rights and explicit consent of any individual whose likeness is used." It also commits to honouring removal requests from those depicted, and the moderation policy confirms individuals may request removal of their likeness at any time.
Now the honest qualification, which matters as much as the rules themselves. Moderation operates through "ML-powered scanning tools to detect potentially non-permissible content as it's uploaded or generated", followed by manual review by trained moderators, with an email appeals route. But neither the moderation policy nor the trust and safety page describes any technical process that verifies consent before an avatar is created. The obligation is stated and enforced after the fact through detection and takedown; it is not gated up front. In practical terms, the platform requires consent but does not check it for you, which means the legal exposure for using someone's likeness without permission sits with you, not with HeyGen. That is a normal arrangement in this industry, but it should be understood rather than assumed away.
The flagship model is Avatar V, described as the most lifelike avatars in AI video. Users can pick from a stock library of over 500 digital twins or create custom ones, with the number of custom twins limited by plan. This is the product's core unit and where its quality reputation is made or lost.
Content can start from several places: Text to Video, Image to Video, Audio to Video, PPT/PDF to Video, Voice Cloning, Voice Director and Gesture Control. The slide-deck route is particularly relevant for corporate use, since turning existing training material into narrated video removes the usual bottleneck of recording a presenter.
Video Translation is presented as a first-class product that breaks language barriers in one click, generating a version of a finished video in another language with the avatar's speech adapted. For organisations serving multiple markets, this is often the strongest single argument for the product, since the alternative is re-recording every asset per language.
Video Agent takes an idea to a finished video from a prompt, while AI Studio provides an editing environment for refinement. These reflect a shift from single-clip generation toward an end-to-end production workflow.
An API is offered for developers, and the Business tier adds LMS integrations, which signals that corporate training is a formally supported deployment rather than an incidental use.
Marketing and sales videos are the most common use — product explainers, personalised outreach and social content produced without booking a studio.
Corporate training and enablement is arguably the strongest fit, especially where material changes often. Updating a script and regenerating is dramatically cheaper than re-shooting, and the LMS integrations on Business plans confirm this is a supported path.
Localisation is the highest-leverage use. One source video becomes many language versions without re-recording, which changes the economics of serving international markets.
Internal communications and onboarding suit the tool well, since consistency matters more than cinematic production values and the presenter is usually someone who has consented as an employee.
Product documentation converted from slide decks or PDFs into narrated video is a natural fit given the PPT/PDF input path.
Conversely, this is not the right tool for emotionally nuanced storytelling, for anything requiring an authentic unscripted human presence, or for any use involving a person who has not consented.
Get consent in writing and keep the record. A verbal agreement is worth little if a dispute arises later, and the platform places this responsibility squarely on you.
Do not assume public figures are usable. This is the most common and most damaging misconception, and the moderation policy prohibits it explicitly.
Check the plan before commercial use. Free-tier output is licensed only for personal, non-commercial and internal evaluation purposes, so a video made on the free plan cannot legally run as an advertisement.
Budget in credits and minutes, not just subscription price. Plans differ in credits, maximum video length and number of custom digital twins, and long-form or high-volume work moves you up tiers quickly.
Script pronunciation deliberately. Brand names, technical terms and non-English words are where synthetic speech most often goes wrong; test them in a short clip first.
Consider the opt-out if your material is sensitive. The privacy policy allows you to request exclusion of your input from model training, which is worth exercising before uploading confidential footage.
Disclose synthetic presenters where it matters. Regulatory expectations around AI-generated likenesses are tightening, and disclosure is increasingly the safer default in advertising and public-facing communication.
Marketing and content teams producing a steady stream of presenter-led video are the core audience, particularly where the same message must exist in several languages.
Learning and development teams benefit most from the update economics and the LMS integrations available on Business plans.
Solo creators and personal brands use custom digital twins to scale their own presence, which is also the cleanest case from a consent perspective since the subject is the account holder.
International businesses gain the most from Video Translation, where the alternative is re-recording every asset per market.
Developers embedding video generation into their own products are served by the API.
It is a poor fit for anyone wanting to use another person's likeness without permission, for productions requiring genuine human performance and emotional range, for free-tier users hoping to publish commercially, and for organisations whose confidentiality rules conflict with uploading facial or voice data to a third-party service.
HeyGen is delivered as a browser-based application, with app.heygen.com as the working entry point and heygen.com carrying the marketing, pricing and policy pages. Generation runs on HeyGen's infrastructure, so no local rendering hardware is required.
The developer API extends the product beyond the interface, allowing video generation to be embedded in other systems — the route most relevant to teams producing videos programmatically at volume.
Enterprise-oriented platform features appear on the Business tier and above, including SAML/SSO and LMS integrations, with Enterprise adding enterprise-grade security and privacy arrangements. The footer also exposes a security portal, a trust and safety page, a moderation policy and GDPR compliance documentation, which is a fuller set of governance material than most tools in this category publish.
Pricing is split between individual and business tracks, and the limits that matter most are credits, maximum video length and the number of custom digital twins.
The Free plan costs nothing and provides 3 videos per month, capped at one minute each, watermarked, with 1 Custom Digital Twin and access to 500+ stock twins. It is genuinely useful for evaluation but restricted in a way that matters legally as well as practically, since its output is non-commercial.
On the individual side, Creator at $29 / mo with 600 credits and Pro at $49 / mo with 1,000 credits form the main paid options. Creator raises video length to 30 minutes and removes the watermark; Pro adds 4K export.
For organisations, Business at $149 / mo with 1,500 credits, videos up to 60 mins and 5 Custom Digital Twins adds SAML/SSO and LMS integrations, with additional seats at $20 each. Enterprise is quoted individually and removes the video duration maximum.
The practical guidance is to model your real workload against credits and length limits rather than comparing headline prices, and to note that the number of custom digital twins is often the binding constraint for teams wanting several presenters.
Synthesia is the closest direct competitor, with a similar avatar-and-script model and a strong enterprise training focus.
D-ID and Colossyan occupy adjacent territory, with D-ID known for photo-driven talking heads and Colossyan aimed heavily at workplace learning.
Descript approaches from the editing side and suits teams whose presenters are real people, offering AI assistance rather than full synthesis.
Runway and similar generative video tools solve a different problem — generating footage rather than a presenter delivering a script — and are complements rather than substitutes.
Simply recording a person remains the right answer when authenticity is the point, and it is worth stating that plainly rather than assuming synthesis is always an upgrade.
The consent requirement is the most important constraint and is covered in full above; the short version is that you may not create an avatar of another person without their explicit consent, public figures included, and the platform requires this without verifying it in advance.
Free-tier licensing is a trap for the unwary. Paid plans give you full rights, but on the free plan you receive only a limited, non-exclusive, non-transferable, revocable licence to output "solely for personal, non-commercial, and internal evaluation purposes", and such output "may not be sold, sublicensed, redistributed, monetized, or used in connection with commercial activities". Publishing a free-plan video as marketing is a licence breach, not merely a watermark inconvenience.
Model training on your input is the default. The privacy policy states that HeyGen may use the input you provide "to train and enhance the models that power our Services", with an opt-out available on request. Since the input here includes facial imagery and voice, this deserves more attention than a generic training clause would.
Synthetic delivery has an expressive ceiling. Avatars have improved markedly, but nuanced emotional performance, spontaneity and genuine rapport remain difficult, and audiences increasingly recognise synthetic presenters.
Costs scale with length and volume. Credits, duration caps and custom twin limits all tighten as usage grows, and long-form content moves you up tiers faster than expected.
Third-party validation is positive but thin: a 4.3 out of 5 rating from 70 reviews on Product Hunt, a sample small enough that it should be cited with its size. The homepage references over a thousand reviews without naming a platform, and no rating from an enterprise review site could be independently verified here. No major technology publication's independent review or compliance investigation was located either, so the governance claims in this entry rest on the vendor's own published policies.
Finally, disclosure norms are shifting. Regulatory and platform expectations around labelling AI-generated likenesses continue to tighten, and what is permitted today may require explicit disclosure tomorrow.
This section carries unusual weight because the data involved is biometric in nature.
The privacy policy is explicit about scope, covering name, email and password, payment information, and — most significantly — face imagery and "parts of faces or bodies within a video or photo", alongside voice, scripts, images and videos you provide. Uploading a video of yourself to build a digital twin means submitting facial and vocal biometric data, and that should be a conscious decision.
Training use is the default with an opt-out rather than the reverse. The policy states that HeyGen may use your input "to train and enhance the models that power our Services", while also providing that you may request to opt out by contacting them. The rights section reinforces this with an explicit entitlement to "opt out of your information being used to train our models". For anyone handling confidential material or the likeness of a consenting third party, exercising that opt-out before upload is the prudent order of operations.
Deletion behaviour is specified concretely: after account deletion, data is "kept in the backups for the purpose of disaster recovery for 60 days" and then automatically and permanently erased. A stated retention window is better disclosure than vague assurances.
Rights coverage spans access, deletion, correction and portability, along with withdrawal of consent and objection to processing and marketing, consistent with GDPR and CCPA expectations. A separate security portal is referenced for the subprocessor list rather than naming them inline.
The governance surface overall — a moderation policy, a trust and safety page, a security portal and GDPR documentation — is more substantial than most tools in this category provide, which is a genuine point in its favour even though consent verification remains user-side.
Only with their explicit consent. The moderation policy requires you to obtain the explicit consent of the individual being represented, and states that without it, creating avatars of other individuals is strictly prohibited. This applies to colleagues, clients and anyone else. The platform does not verify consent before creation, so the legal responsibility rests with you.
Explicitly prohibited without their consent. The policy names celebrities and public figures directly. Public availability of someone's image does not create permission to synthesise them.
On paid plans, yes — you own all rights in your input and output. On the free plan, no: output is licensed solely for personal, non-commercial and internal evaluation purposes and may not be sold, sublicensed, redistributed, monetised or used in commercial activities. Check your plan before publishing anything commercially.
The free plan gives 3 one-minute watermarked videos per month. Creator is $29 per month with 600 credits and 30-minute videos, Pro is $49 with 1,000 credits and 4K export, Business is $149 with 1,500 credits, 60-minute videos and 5 custom digital twins plus $20 per additional seat, and Enterprise is quoted individually with no duration maximum.
By default it may. The privacy policy states input may be used to train and enhance the models powering the service, and separately grants you the right to opt out of your information being used to train the models, by request. Given that the input includes facial and voice data, consider opting out before uploading sensitive material.
Data is retained in backups for disaster recovery for 60 days after deletion and then automatically and permanently erased. Access, deletion, correction and portability rights are available under the policy.
Yes. The moderation policy states individuals may request removal of their likeness from the services at any time, and the trust and safety page requires users to honour removal requests from those depicted.
Through a two-stage process: machine learning scanning detects potentially non-permissible content as it is uploaded or generated, followed by manual review by trained moderators. Moderation decisions can be appealed by email. Note that this is detection after creation rather than consent verification before it.
Video Translation is presented as a one-click way to produce another-language version of a finished video and is one of the product's strongest differentiators for international teams. As with any automated localisation, output should be reviewed by a fluent speaker before publication, particularly for idiom and product terminology.
The enterprise surface is reasonably complete: Business adds SAML/SSO and LMS integrations, Enterprise adds enterprise-grade security and privacy, and the company publishes a security portal, moderation policy, trust and safety page and GDPR documentation. The remaining question for procurement is the biometric data and training-by-default posture, where the opt-out should be arranged in writing before deployment.
