Hedra

By Minel Gunesoglu, founder. I divided Hedra's own published credit allowances by its own published credit rate, quoted its pricing page, Terms and Privacy Policy and read its acceptable-use rules, dating each by when it was read or last revised, and logged which review platforms refused to open. Nothing here comes from a video I generated, because I generated none. Last updated 12 August 2026; re-checked monthly.

The most useful sentence published about Hedra was published by Hedra. In a help-center post dated 4 December 2025, the company's own writer states the limit a demo reel is built to hide: "Consistency degrades over time. A 4-second clip typically holds together better than a 10-second clip from the same prompt" (Hedra blog, Erna Balayan, 4 December 2025). That is the vendor, in its own help content, telling a paying customer the long take is the hard case.

Hedra is the vendor at hedra.com, and Character-3 is the model it currently sells: a still photograph plus an audio track or a script goes in, and a talking, lip-synced video of that face comes out. Elements is Hedra's name for the saved character identity meant to persist across separate generations. Composer, its timeline editor, assembles several clips into something longer. Those four names are not interchangeable and this page keeps them apart, because nearly every claim below belongs to exactly one of them. A fifth set of names sits inside Hedra's credit menu and belongs to other companies entirely; the pricing section names them once, and this page scores none of them.

Everything that follows runs into the same obstacle, and it is better named before the first table than after it. Almost nothing about Hedra's output has been measured in public by anyone who does not sell a competing product.

Four Rows of Four: What Hedra's Score Card Verifies, and the One That Changed Its Name

Every cell on this card now holds a figure, and one of them is not the thing its position implies. Two come from documents Hedra publishes about itself (the price list, which converts into a value figure, and the Terms of Service, which answer the questions the commercial-terms cell is built from. The third needs somebody outside the company and took three attempts to get: a written-review aggregate, finally read on 17 August 2026 in a rendered browser after every scripted fetch had been refused. The fourth needed an independent arena, and none exists for avatar output anywhere), checked by name on 1 and 14 August 2026 and again on 19 August. So on 19 August 2026 that row stopped waiting and started publishing a different measurement, under the name of what it measures.

Four cells multiply out to 22,083,665; its fourth root is 68.55, published on the card as 68.6 on the hundred-point scale and 3.4 on the five. Read what the top row is before reading the total. Delivery envelope asks what a paying customer can walk away with; it does not ask whether the video is any good, and a vendor shipping poor avatars through a clean, well-documented pipeline would score high on it. So 68.6 rates a customer record, a price list, a contract and an exit. It does not rate a video, and no sentence on this page will pretend otherwise. It is also not a mean over the four dimensions this site names, since one of the four is answering a different question: set it against a card whose rows all sit on their own axes and the two numbers are not the same kind of object.

That 68.6 is the fourth figure this card has carried, and every move was ours rather than Hedra's. It published 56.0 until 12 August 2026, 77.6 until 17 August 2026 and 65.6 until 19 August 2026, and the 17 August correction is the one worth reading, because it went the other way. Usability had shipped empty on the stated ground that no retrievable review aggregate existed. That ground was wrong, and the reason was method rather than posture: every attempt behind it was made with a scripted fetch, and Trustpilot refuses those from every host it serves. Hedra was never refused by a browser. It was never shown to one. A rendered browser on the New Zealand mirror returned 1.7 out of 5 across 53 reviews on 17 August 2026. Worse, a crowd sweep had already read that same figure on 14 August and it was never filed. Filling the cell cost Hedra twelve points, which is the part that matters: a blank is not neutral, and the missing row here was the weak one. The fourth cell used to be blank on the stated ground that no indemnification clause had been located in the Terms. Section 8 of that document is headed Indemnification, in a text whose own last-revision stamp reads 25 October 2024. Our note was wrong, and a note recording an absence for a reason that was never true cannot be caught by re-reading the note. Reading the document filled the cell, and because the cell came in above the value figure, the composite rose. A blank that flatters and a blank that penalises look identical on a card, which is the argument for going back to the source every time.

The 19 August move is a different kind of event and should not be read as another correction. Nothing was found to be wrong. The site stopped publishing empty rows, the last blank here took a delivery-envelope reading of 78.1, and because 78.1 sits above the 65.6 the card was carrying, the total went up by three points. Hedra's product had no part in that. A reader who wants the movement in one sentence: the card got fuller, not better.

A card carrying a renamed row invites a fair objection, and the answer is that the rename is the finding. Read together, three axes and one substitution say something specific and checkable: as of 19 August 2026 the public record on this product supports a judgement about what Hedra charges, what its contract commits it to and what a subscriber can carry out of the product, and no judgement at all about what Hedra produces. Other Hedra reviews do publish numbers, and one publishes more of them than this page does, a nine-dimension card rolling up to 3.90 out of 5. Of the seven Hedra reviews read in full while assembling this page on 1 August 2026, not one states where a dimension's evidence came from, how large the sample was, or on what date it was collected. The nine-dimension card is the clearest case: beside its dimensions it also prints an uptime figure of about 99.5 per cent and a monthly-active-user figure above 100,000, and neither number appears anywhere on hedra.com. A precise-looking figure and a sourced one are different objects, and the difference is not visible to a reader who does not go looking. The rule that decides when a cell may carry a criterion under another name, and what the card must print beside it, lives at what each row on a card is allowed to mean, and it applies the same way to every tool this site scores.

Hedra was the first tool to receive a card in the avatar-generator category on this site, and since 14 August 2026 it has a sibling there, so 68.6 can now be read against something. Two of those rows do compare cleanly: both cards carry the same delivery-envelope criterion in the capability position, read off each vendor's own documents with the same rubric, so 78.1 against the sibling's figure is a like-for-like reading. The rest wants care. The sibling's usability rests on a far deeper sample than this one, and a difference in what could be measured is not a measured difference between the products.

Why No Arena Covers Hedra Character-3's Capability

Capability on the other cards published here converts a blind-vote arena rating: a voter is shown two unlabelled outputs of one prompt and picks the better, and enough of those choices resolve into a public ladder anyone can open. Artificial Analysis, the board this site reads for video models, was checked by name for both "Hedra" and "Character-3" on 1 August 2026 and carries neither. No substitute board covering avatar or lip-sync models surfaced across the seven Hedra reviews read the same day. Checking the source against the tool before the score is built, rather than after, is the whole reason that cell was never filled with a rating borrowed from a board measuring something else.

What sits there instead answers a smaller question honestly. Delivery envelope reads three positions off Hedra's own documents. The avatar models' published resolution list (an "Output resolution." enum of 540p, 720p and 1080p on both Character-3 and the avatar endpoint), puts the ceiling at 1080p. Hedra's pricing page states that plans from $15 a month include commercial use with no watermark, while a cancelled account keeps "limited watermarked generations", so the mark is on the unpaid state rather than the paid one. And the API is a real public contract: 94 paths in an OpenAPI 3.1 document the vendor's own docs index links, readable without an account or a key.

One reading was refused in the direction that cost Hedra points, and the refusal is worth stating because the opposite would have been easy. Hedra sells 4K on that same $15 tier. It is an upscale, the vendor's page credits Topaz, bills three credits per second of source footage at 1080p and twelve at 2K and 4K, and says in its own words that "Most video models output 720p or 1080p" (hedra.com/uses/ai-video-upscaler, read 19 August 2026). An upscale is a second operation on a clip the avatar model has already capped at 1080p, so counting it here would turn this row into a measure of whether a vendor resells an upscaler. The ceiling scored is the generation's own. Two facts travel with it and neither is optional: the resolution list is API-wide rather than tied to a named plan, and the API bills from a wallet Hedra states is separate from the Studio balance, so pipeline access is a second bill rather than something the $15 includes.

One figure circulates widely enough to be worth naming and refusing. Several Hedra write-ups report that independent evaluation puts its lip-sync accuracy at around nine out of ten. None of them names who performed the evaluation, on what material, or on what date. A figure with no author is not evidence that a tool is good; it is evidence that a sentence was copied. It does not enter the card above, and it does not enter it as a hedge either.

Hedra Lip-Sync Over a Long Take and Identity Across Cuts: the Vendor's Own December 2025 Admission

Hedra's help-center post does not stop at conceding that consistency degrades. It tells the reader what to do instead, and the instruction is the part worth paying attention to: "Generate multiple short, high-quality clips and edit them together" (Hedra blog, Erna Balayan, 4 December 2025). Set that beside a landing-page demo and the two describe different products. The demo implies you hand Character-3 a script and receive a performance. The documentation says you hand Character-3 several short prompts and assemble the performance yourself.

Note carefully what the admission covers. Erna Balayan's example weighs a 4-second clip against a 10-second one from the same prompt, which is degradation inside a single generation. Identity across cuts is a different question: whether the same face, hair, wardrobe and lighting return when segment two is generated an hour after segment one. Hedra's answer to that question is Elements, the saved-identity feature, and nothing Hedra publishes quantifies how well Elements holds. Nothing found in the public record on 1 August 2026 measures it either.

Two other publishers have observations close enough to report as theirs. techpoint.africa ran one identical prompt four times and rejected two of the four results, noting a prop present in one output and absent from another (published 1 September 2025). aimagic.fun reports regenerating roughly one output in eight because of hand and pose artifacts (updated 18 May 2026). Both describe variance between separate short generations from a static prompt, which is not the same measurement as identity holding across an edited sequence, and neither observation is this desk's. No controlled repeat run stands behind any sentence on this page, which is exactly why neither rate is converted into a figure on the card above.

Hedra's Composer Timeline Editor: Does Sync Survive the Stitch Between Clips?

Composer is Hedra's timeline editor, and its job is the workaround the December 2025 post recommends: assemble several generated clips into one longer piece. The seam is therefore the thing to ask about before paying. Inside a single Character-3 clip the vendor has already told you consistency falls off with length. Between two clips the questions are different: whether the voice keeps its level, whether the face returns under the same lighting, whether the mouth picks up mid-word where the previous clip stopped. Hedra publishes no statement on any of the three, and no dated account located on 1 August 2026 answers them. Anyone whose deliverable is a two-minute talking-head piece is buying that unknown rather than avoiding it. The one move available before committing is to generate two short segments on the free entry and cut them together yourself, because the seam is the part no demo reel shows.

Trustpilot, G2, Product Hunt and Reddit: Why Hedra's Community Trail Goes Cold

Hedra's public story is scale. Coverage in 2026 cites more than 20 million users and a $32 million Series A led by a16z; neither figure appears on hedra.com's own pages as checked on 1 August 2026, so both are carried here as press and vendor-sourced claims rather than as anything verified independently. The one scale figure Hedra publishes itself counts something else again. The homepage sets "125k+" above the line "businesses have used Hedra inference and infrastructure" (hedra.com, captured 12 August 2026), a count of businesses that have touched the inference and infrastructure layer, not of users, and it cannot be reconciled with the press number without knowing how either was arrived at. A correction attaches to the wording rather than the figure: an earlier version of this page put Used By Over 125k Businesses in quotation marks against a 1 August 2026 read, no capture from that date was kept, and the sentence the homepage carries is the one quoted here. A product at that scale would be expected to leave a wide public trail behind it. Hedra's is close to invisible, and the sweep that found the absence is the same sweep that produced two of the blanks on the card above.

Platform by platform, on 1 August 2026: Trustpilot refused four direct requests across two addresses. G2 refused as well. Capterra carries no listing. No dedicated subreddit exists, and general searching returned nothing verifiable across roughly ten phrasings. Product Hunt was the one platform that opened, showing four launches between June 2024 and March 2025 carrying 165, 149, 2 and 145 upvotes, with no written review on any of them.

Those five are review platforms, and the sentence above claims nothing beyond them. Hedra does run channels of its own, linked from its own homepage footer and read there on 1 August 2026: a Discord server, a LinkedIn page and an Instagram account. What is said inside them was not audited for this page, and a vendor-run room is a different kind of evidence from an independent review anyway. People arrive there already using the product, and the host can moderate. The distinction that matters for a buyer is narrower than it first looks. What is missing is not conversation about Hedra. What is missing is conversation a prospective customer can read without the vendor standing between them and it.

That last count needs a correction attached to it, because a quotation that does not exist is circulating as evidence. Search summaries repeatedly surfaced a specific five-star Product Hunt review of Hedra, worded I have so much fun using Hedra! It's so much fun and rather intuitive to use.: italic, not quoted, because on this evidence nobody is on record saying it. Fetching Hedra's Product Hunt reviews page directly, and the individual review permalink those summaries pointed at, returned "No reviews yet" both times on 1 August 2026. The upvotes are real and countable. The quotation, on that check, is not on the page it is attributed to.

Trustpilot is the platform that would have mattered most, and its refusal is why the usability cell above is blank rather than estimated. A star rating for Hedra does circulate, and the circulating versions disagree with each other: 2.1 out of 5, 1.9, 2.0 across 32 reviews, and 34 reviews at 76 percent one-star. Those are summaries of a page nobody in this sweep could open. Four different figures for one page is not a rating with a margin of error, it is a sign that a number is being passed along rather than read, and none of those four figures is averaged, adopted or scored here.

Until 12 August 2026 that same wall cost a second cell. The fourth dimension used to average a consumer star rating with a reading of the contract, so a platform serving a bot check emptied half of it and the whole cell went with it. It does not any more: what the 81.3 converts is Hedra's own Terms of Service and its own pricing page, neither of which decides who may read them on a given afternoon. No figure from Trustpilot, G2 or Product Hunt stands behind that number, which is why it exists at all.

The record that does exist is three Hacker News threads from 2024, and it is worth reading precisely because nobody in them set out to review anything. On a rival product's launch thread on 3 August 2024, a commenter volunteered that "Hedra is best for lipsyncing" (Hacker News). On 6 September 2024 another wrote, "We are big fans of Hedra. Do you know if they've publicly commented on their model architecture?" (Hacker News), and in a separate thread the same day a third warned a competing founder, "You should watch out for Hedra and Sync. Plus a bunch of Loopy activity on Discord." (Hacker News). Three unprompted mentions from people with no visible tie to the company is a thin sample from a single platform. What makes it worth carrying is the date on it: all three predate Character-3, so the standing being described belonged to an earlier model than the one credits buy today.

The same September thread carries the sharpest published criticism of Hedra's product, and Hedra answered it on the record. Told that "you've got one video on your homepage and everything else is locked behind a signup button" (Hacker News, 7 September 2024), a Hedra team member replied, "We have a sign-up because we ensure users accept our terms of service and acceptable use policy before creating their first video" (Hacker News, 7 September 2024). Whether that is the whole reason is not settleable from a comment thread, and other commenters in it were unconvinced. What the exchange does establish, dated and attributable, is that the signup wall is where agreement to the licence terms further down this page gets collected.

One absence deserves stating rather than glossing. The billing and cancellation complaint pattern that dominates the community record of nearly every tool scored on this site could not be checked for Hedra at all, because that pattern lives on the two platforms that refused every request. No complaint record here is a hole in the evidence, not a clean bill of health.

Hedra AI Pricing: 1,500 Credits for $15, and the Menu That Also Sells Kling and Veo

Hedra sells seconds, and prices them one at a time. Every row on the ladder below converts through a single published rate: Character-3 costs 6 credits per second of output, flat, with no resolution split stated anywhere on the page (hedra.com/pricing, retrieved 1 August 2026). One third-party guide reports 3 credits per second at 540p against 6 at 720p; Hedra's own page states a single number, so a single number is what this review uses.

Prices and credit counts come from hedra.com/pricing on 1 August 2026. The last two columns are division and nothing else: credits divided by six is seconds, and the arithmetic can be redone on the vendor's own page in less time than this paragraph takes to read. That is the only reason those columns can be published at all.

The Free row is a genuine gap rather than an oversight on this page. Hedra's pricing page offers a "Start free" entry and attaches no credit allowance to it. Third-party reviews fill that silence with a figure and disagree about which figure: some say 100 credits, others say 300, and none says where it came from. At 6 credits a second the gap between those answers is roughly 17 seconds of output against 50, which is the difference between being able to evaluate the tool and not. No number ships here, because Hedra's own pricing page and help centre, both read 1 August 2026, state none.

Two entry terms sit on either side of that silence, and the friendlier one is printed at the top of the same page: "Free Plan, No Credit Card" (hedra.com/pricing, retrieved 1 August 2026). Evaluating Character-3 therefore costs no payment details up front. The second term is where the vendor's wording runs out. The only watermark language anywhere on that pricing page attaches to cancelled accounts, which keep "limited watermarked generations" after losing everything else; what marking, if any, a free-plan generation carries is not stated there. That silence lands directly on the one instruction this review gives, because a free take stands in for a paid one only if it comes back the way a paid one would. Settle it on the plan card or with Hedra's support before reading a free result as a preview of what $30 buys.

Alongside Character-3, Hedra's credit menu also sells generations from other companies' models, including Kling 2.5 Turbo, Veo 3.1 Fast and Sora 2 Pro. Hedra also sells a separate real-time streaming product, Live Avatar, priced separately from the credit subscription; nothing about how it is priced is quoted here, because no dated vendor surface stating it was found.

Two of the Hedra reviews still circulating quote a price list Hedra no longer sells: a Lite tier at $8 for 1,000 credits, Creator at $24 for 4,000, Professional at $60 for 12,000. Those pages carry publication dates of 1 September 2025 and 1 November 2025, and neither has caught up with what changed since: the two tiers that kept their names both rose exactly 25 percent (Creator from $24 to $30, Professional from $60 to $75), while the $8 Lite tier was replaced by Basic at $15, which is a rename rather than a comparable increase. If a Hedra price you read anywhere carries no date, including on this page, treat it as unverified until you have checked it against the vendor's own page.

Hedra Creator Plan: 5,400 Credits Is 15 Minutes of Character-3, Not 30

The $30 tier is where most of this decision gets made, so here is the whole calculation in one place. Creator provides 5,400 credits a month. Character-3 consumes 6 credits for every second of output. 5,400 divided by 6 is 900 seconds, which is 15 minutes of finished video a month, not the 30 that a dollar-a-minute reading of the price tag suggests. That division is not unpublished: at least one other review already puts Creator at roughly fifteen one-minute videos a month, and this page does not claim to have found it first. Two things here are its own. The first is that the numbers going into the division were re-read from the vendor's page on 1 August 2026, while two of the reviews carrying Hedra figures are still quoting a tier that was renamed and a price that rose. The second arrives in the next paragraph, and it is the part none of the seven reviews read for this page finishes. The same division runs the rest of the ladder: Basic's 1,500 credits is 250 seconds, 4 minutes and 10 seconds; Professional's 14,400 is 2,400 seconds, 40 minutes.

Then start subtracting, because 15 minutes is the ceiling before anything is thrown away. A discarded take bills exactly like a kept one, since credits are spent on generation rather than on approval. One 30-second attempt you decide against costs 180 credits, which on Basic is about an eighth of the month and on Creator about a thirtieth. This page asserts no discard rate for Hedra, because no controlled run stands behind one. What it can do is carry the two figures other publishers put near this problem without ever completing it: techpoint.africa rejected two of four identical attempts (1 September 2025), and aimagic.fun reports regenerating about one in eight (18 May 2026). Apply the gentler of those to Creator's 15 minutes and roughly 13 usable minutes remain; apply the harsher and the same $30 buys about seven and a half. Neither rate is Hedra's and neither is this desk's, and the spread between them is the honest answer: the figure that decides this purchase is minutes that survive review, not dollars on a plan card.

Hedra does not publish three things the arithmetic above would need, and the surfaces checked on 1 August 2026 were the pricing page, the documentation overview and the homepage. No output resolution for Character-3 appears on any of the three, and one surface not checked that day does carry it, which is the correction this paragraph now has to make. The API contract read on 12 August 2026 lists an "Output resolution." enum of 540p, 720p and 1080p for the avatar models, and that is the ceiling the score card's top row reads. It does not rescue the credit arithmetic below, because the enum belongs to the API rather than to a named plan, and no captured document ties a resolution to Basic or Creator. Whether reaching a higher resolution costs additional credits is still stated nowhere for a subscription; the upscaler that reaches 4K is priced separately, in credits per second of source footage. Neither is a generation time or a queue estimate. Read the minutes accordingly. Six credits a second is a rate of consumption, so dividing an allowance by it answers how much video a plan will generate and nothing beyond that. Should a higher resolution bill at its own rate, every minute figure on this page describes a cheaper default rather than the file a client receives, and Hedra publishes nothing that would settle which. What a plan buys, on the vendor's own published numbers, is generation time. Whether the minutes that survive review also arrive at delivery quality is a question the pricing page never reaches.

Hedra Monthly Credits Rollover and Refunds: a Completed-but-Drifting Generation Still Bills You

Rollover is the second subtraction, and Hedra states it four times on one page. "Monthly Credits Do Not Roll Over" appears on Basic, Creator, Professional and Teams, and the pricing page's own FAQ repeats that subscription credits "do not carry over between billing cycles" (hedra.com/pricing, retrieved 1 August 2026). Credits bought separately as a pack behave differently and do not expire. A quiet month is therefore forfeited rather than banked, so the plan sized to an average month will underserve a busy one and waste a slow one.

Refunds are where the record stops being clean, and this page is going to leave it that way. Hedra's billing documentation states that a failed request is refunded automatically (hedra.com/docs, read 1 August 2026), with the net credit charge landing at zero, which reads as a cap on the downside of a bad generation. Set beside it the distinction that documentation does not draw. A generation that completes and returns a video whose lip-sync has drifted off the audio is, technically, a success. Whether that outcome carries any refund route is stated in nothing Hedra publishes and could not be verified on 1 August 2026, and the platforms where a paying subscriber would have complained about it are the two that refused every request. Offered as open reasoning rather than as a finding: the space between a technical success and a usable video is where cost-per-minute pain would sit, and the public record neither confirms nor closes it.

Hedra Terms of Service and Likeness Consent: What Happens to Your Avatar If You Cancel

Two clauses in Hedra's Terms of Service, both revised 25 October 2024, have to be read together rather than one at a time. The first grants the company a great deal. Uploading content gives "Hedra a non-exclusive, transferable, perpetual, irrevocable, worldwide, fully-paid, royalty-free, sublicensable right and license to use, copy, reproduce" it (Hedra Terms of Service, 25 October 2024). The second gives something back: "Hedra does not claim ownership of any Inputs or Outputs", which the same sentence then defines collectively (in the source's own double quotation marks), as "Your Content" (same document, same date). A small correction rides with it: an earlier version of this page set that defined term in single quotes inside the quotation, and the characters inside a quotation belong to the source. Both hold at once, and they are not quite the contradiction they first look like. You keep ownership, and Hedra keeps a permanent, sublicensable licence to use what you uploaded. For a photograph of your own face those are two very different reassurances, and neither clause resolves the distance between them. A third line about commercial rights sits on the pricing page rather than in either clause: "Commercial use" appears there as a paid-plan feature (hedra.com/pricing, retrieved 1 August 2026), and what a free-plan generation may be used for commercially is not stated.

A third clause in the same document is the one carrying most of the 81.3 above, and it is the clause a page describing what somebody is buying should quote rather than summarise. Section 8, headed Indemnification, obliges the customer: "You shall indemnify and hold Hedra ... harmless", read in full on 12 August 2026, and nothing anywhere in the document runs in the other direction. Practically, a Hedra generation that draws a third-party claim is a claim you answer on your own. That is the ordinary direction of a software contract and not a peculiarity of this vendor, which is why the cell reads in the eighties rather than at the floor: the ownership sentence quoted above is the other half, and on the tier a subscription buys it says outright that the output is yours. Of the four positions this dimension can hold, those two were answered. What a free-plan generation may be used for commercially is answered by no Hedra document, so it was left out of the average rather than guessed at, and the refund position was not read in this pass and is recorded as unread rather than as absent.

Hedra's Acceptable Use Policy governs the other direction, and it is the part to know before uploading anybody else's photograph. The face has to be your own or carry explicit written consent from the person shown, and non-consensual deepfakes of identifiable people are prohibited outright (acceptable-use rules, read 1 August 2026), with a carve-out for historical figures dead more than a century. Recorded here as a consumer-protection fact, not as a comparison of whose consent gate is stricter. On Hedra's own account, given publicly in 2024, the signup wall is where agreement to that policy is collected before a first video exists.

Hedra Cancel Subscription: Watermarked-Only Access Is Confirmed, Your Element Data Isn't Addressed

Cancellation is documented, narrowly. Hedra's pricing page states that a cancelled account "will lose access to most functionality of the platform, apart from limited watermarked generations" (retrieved 1 August 2026). The account survives; the usable output quality does not.

What neither governing document addresses is the question people actually ask about a tool built on their own face. Whether the uploaded photograph is deleted, whether the saved Element identity built from it is destroyed or merely locked, and on what timetable either happens, is stated in neither the Terms of Service nor the Privacy Policy, both revised 25 October 2024. The Privacy Policy's retention language is generic: "When we no longer require the personal information we have collected about you, we may either delete it, anonymize it, or isolate it" (Hedra Privacy Policy, 25 October 2024). No trigger, no timetable, and no clause specific to biometric-style data. Read against the perpetual licence quoted above, the honest summary is that cancelling stops the charge and settles nothing about the face.

Hedra's Alternatives Are Named Here, Not Ranked: HeyGen, Synthesia and Runway

A Hedra shortlist almost always contains HeyGen, Synthesia and Runway, and what this page can honestly do with those names is narrow: say what each one gets reached for. Which of them carry a card here moves as tools are scored, so this section links the ones that do rather than counting them. Synthesia, HeyGen and D-ID are all scored on this site, and their composites are not printed in this section, for the reason the next sentence gives. Runway is not scored here, so a figure for it would have to be invented. Ranking any of these against Hedra fares no better. Against an unscored name it would set a real measurement against an invented one; against a scored one it would set two cards whose cells rest on different evidence, and the difference there is which review platforms opened rather than which product is better. HeyGen is the name raised most often beside Hedra, in the same photo-to-talking-avatar shape. Synthesia comes up where the work is corporate training and localisation rather than creator output. Runway comes up where avatar work is one job inside a general video toolkit. Each of those three sentences records where a name tends to come up. None of them is a finding about output, and none is scored here.

One caution about where the comparison gets read matters more in this category than in most. Search for a Hedra review and roughly half of the pages that come back are published by a company selling a directly competing avatar or video generator, with no disclosure anywhere on the page. On two independent search indexes read on 1 August 2026, both of the first two results returned were that kind of page: each is written as a review and each closes by recommending its publisher's own product, one of them under a heading offering its tool as the answer to Hedra being unavailable. None is named here, because naming them promotes them. A duller check does as much work as any of this: look at the date, and at which model is being described. Tom's Guide's three Hedra posts are all dated 2024, and the main one covers Character-1, two model generations before the one credits buy today. On any comparison you read next, this page included, check the byline and check who owns the product recommended at the end. No affiliate relationship exists between this site and Hedra or any of the three names above.

Is Hedra Worth $30 a Month? The Verdict for a Solo Creator's Highlight Reel

Dated 1 August 2026, and organised around the two readers this evidence can actually serve.

For a solo creator making short talking-head pieces, a channel intro, a product explainer cut into segments, a run of social clips from one script, the arithmetic favours Creator and argues against Basic. Carry the adjusted figure into that decision rather than the headline one. Creator's fifteen raw minutes land between roughly seven and a half and thirteen usable ones once discards are paid for, on the two outside rates quoted earlier, and a weekly two-minute piece needs eight. At the gentler rate that leaves room for a spare take or two across the month; at the harsher one the same $30 falls just short of four two-minute pieces. Which end you land on is set by your own discard rate rather than by Hedra's price list, and it is the first thing worth finding out. Basic's four minutes and ten seconds covers little beyond that discovery, which is arguably what an entry tier is for. Start on the free entry, which the pricing page offers with no card required, and pay only once your own face has come back looking like your own face and you know whether a free take arrives marked.

For anyone arriving here having been burned by a subscription before, four dated facts on this page should decide it. Credits do not roll over. A discarded take bills like a kept one. Cancelling leaves watermarked output and settles nothing about the photograph already uploaded. And the two platforms where subscribers usually log billing complaints refused every request made on 1 August 2026, so the quiet around Hedra's billing record is missing evidence rather than good news.

Do not buy Hedra this month if the deliverable is one continuous multi-minute performance. The vendor's own December 2025 post says to build it from short clips and edit them together, and nothing public establishes whether the seams hold. Do not buy on the strength of a star rating whose source you have not opened, including the two-point-something average circulating for this product, which nobody in this sweep could reach. And do not read the number at the top of this page as a verdict on the videos.

Take the 68.6 for what it measures: a customer record, a price list, a contract and an exit, each checked against its own source on the day it was read. The part of this review that should decide a $15 or $30 charge is the division, 1,500 over 6 and 5,400 over 6, because it can be redone on hedra.com in under a minute, and it does not require believing a word this page says. The renamed row is not a criticism of Hedra's output and it is not a compliment either. It records that nobody, this desk included, has published a measurement of the videos, so the card publishes the nearest thing that traces to bytes and labels it as what it is.

Minel Gunesoglu, founder of Vouch, owns the division above and the renamed row beside it · LinkedIn · what this desk is allowed to claim. Hedra data collected 1 August 2026; every price, credit figure and quoted clause above carries a date, either of the reading or of the document's own last revision, and all of them are re-checked monthly. Disclosure: no affiliate relationship exists with Hedra, so every vendor link here is a plain link; where affiliate links appear elsewhere on this site they are marked, and a paid submission buys review speed only, never a listing and never a score.