Published Updated
In one or two sentences, before anything else.
Hedra is an AI company whose Character 3 and Avatar models turn a portrait and an audio track into talking video up to 10 minutes long, sold through its Studio agent and a pay-per-use API. In late September 2026 its US$30 Creator plan became Pro at US$50 a month. Plan credits reset monthly; no free plan is listed.
What it can do, one sentence each, and whether we actually tested it.
Turn a start image and an audio track into a talking or singing video of up to 10 minutes, at 540p, 720p or 1080p in seven aspect ratios from 21:9 to 9:21, with speaker selection. The length follows the audio.
Per Hedra's model pages. Hedra dates Character 3 to March 2025; the Avatar page says it runs on Together AI infrastructure. Both list the same API rate: 2.5 cents a second at 540p, 5 at 720p, 6.25 at 1080p.
You brief an agent in a Space and it plans, generates and edits images, video and audio with the models it picks; Elements keep characters and products consistent, Skills save repeated workflows, and teammates can work in the same Space.
Per Hedra's Studio page and help center. Hedra's own benchmark of the agent, published March 31, 2026, reports a task success rate of roughly 60%.
Runs Kling V3, Veo 3.1, Seedance 2.5, MiniMax H3, Wan 3.0, Nano Banana, GPT Image 2, ElevenLabs voices and rival avatar models such as HeyGen Photo Avatar 4, OmniHuman, Kling AI Avatar v2 and VEED Fabric, all on one account.
Per Hedra's live API catalog on November 3, 2026: 110 models, 53 of them video, 40 image, 12 audio and 5 text. Its trust center lists fal.ai, BytePlus, MiniMax, ElevenLabs, Kling AI and HeyGen among the providers.
Besides the agent, Studio has direct Image, Video (all, avatar, transition and upscale modes), Speech and Edit Video tools; Edit Video, also called Composer, arranges clips, images and audio on a timeline and exports one file.
Per Hedra's help center. Credit costs per model show only in the app's live selector, which the help pages say they do not repeat.
Generates voiceovers from a script with ElevenLabs and MiniMax voices and clones a voice from a recording; the audio can drive an avatar video in the same Space.
Per Hedra's pricing page (voice cloning free on the API, speech from 5 cents per 1,000 characters) and its voice cloning page, which says to clone only a voice you have permission to clone.
One API key reaches every catalog model through typed asynchronous jobs, with a free cost-estimate call, signed webhooks, Python and TypeScript SDKs, a CLI, and OpenAI- and Anthropic-compatible endpoints for the text models.
Per Hedra's developer API post of October 6, 2026. The wallet starts at US$0 and Studio credits cannot pay for API calls.
Adding mcp.hedra.com to Claude, ChatGPT, Cursor, Claude Code or Codex lets the assistant list models, estimate a job, generate media and save it to your Hedra library after you sign in.
Per Hedra's MCP docs. MCP jobs can be paid from a Studio subscription or the API wallet, the one exception to the rule that Studio credits do not fund the API.
Teams costs US$75 a month for 14,400 credits with the fastest generation and team access; Enterprise adds custom credit volumes, single sign-on, private deployments and legal and security review.
Per Hedra's pricing page. Its trust center lists a SOC 2 Type I report, available on request.
The people it helps and the problems it solves, and where it ranks.
Built for
Problems it solves
Public tests other people ran: what they gave it and how it did.
Case 1
Input
Midjourney portraits for three rounds (a neutral front-facing face, a wide smile showing teeth, a person turned side-on mid-gesture), each tool's built-in voices and the same monologue, cut to 10 seconds.
What happened
Hedra won the neutral-face round for the most realistic voice and mouth, won the smiling round only as the least bad, and lost the side-on round to Runway, finishing first overall. It animates a still image while the others map lips inside a video. Two planned rounds were dropped because Kling rendered too slowly.
Case 2
Input
60 seconds of 1080p avatar video from one portrait and one voice track, on Hedra Avatar or Character 3, then on rival avatar models Hedra also hosts.
What happened
At 6.25 cents a second, Hedra's own models cost US$3.75 for the minute, against US$6.00 on HeyGen Photo Avatar 4 and US$9.60 on OmniHuman 1.5 at Hedra's listed rates. A 10-minute 1080p video, the longest Character 3 allows, comes to US$37.50, by our arithmetic from the catalog.
Case 3
Input
A 10-second, 1080p clip with native audio from one start frame and a text prompt, one clip per call.
What happened
Prices ran from US$0.60 on LTX-2.3 Fast to US$7.91 on Seedance 2.5, a 13-fold spread for the same output; Kling V3 Pro came to US$1.68 and Wan 3.0 to US$2.00. These are Hedra's API dollar rates; the matching Studio credit costs are not published.
Case 4
Input
A six-second MiniMax H3 clip at 768p, prompt 'a fox sprinting across fresh snow', submitted with the Python SDK after funding the wallet.
What happened
Hedra prices it at US$0.36. Until the prepaid wallet has money in it, every request returns a 402 INSUFFICIENT_BALANCE error, and a Studio subscription does not count.
Case 5
Input
Recorded multi-turn conversations (an image request, a video built from that image, then edits) replayed with simulated media generation after each change to prompts, tools or models.
What happened
Removing a planning mode and injected reminders raised the agent's probability of completing the scenarios from roughly 40% to 60%, by Hedra's own measure. No outside party has published a comparable test of the agent.
What it does badly, with the evidence, and what we could not settle.
Weak
The plans got more expensive in late September 2026.
Evidence: Archived copies of the pricing page show Basic at US$15 for 1,500 credits, Creator at US$30 for 5,400 and Professional at US$75 for 14,400 on September 14, and Basic US$20 for 2,000, Pro US$50 for 7,200 and Ultra US$100 for 18,000 by September 29. A Pro credit costs about 0.69 cents against 0.56 on Creator, by our arithmetic.
What a Studio credit buys is not public.
Evidence: The help center says exact credit prices per model are deliberately left out and shown only in the app's live selector; the public pages publish dollar rates for the API alone.
Credits are use-it-or-lose-it unless you keep paying.
Evidence: Monthly plan credits reset each cycle, annual plans now grant credits month by month, and pack credits, though they never expire, can only be used with an active subscription, per the pricing FAQ and help center.
There is no visible free plan or refund policy.
Evidence: The current pricing page lists no free tier, where it said 'Start free, upgrade anytime' until at least October 7, 2026, and Hedra's model pages say generation uses paid credits. No refund policy is published; the help center only says failed requests may be refunded automatically.
Avatar output stops at 1080p and is never live.
Evidence: Character 3 and Hedra Avatar render at 540p to 1080p, and every API job is asynchronous. LemonSlice, which sells real-time avatars, found no interactive avatar product on hedra.com in September 2026.
Independent evidence is old.
Evidence: The only outlet test found, by Tom's Guide in October 2024, covered the older Character-2; no outlet test of Character 3, Hedra Avatar or the Studio agent turned up, and Trustpilot's page for Hedra was closed behind a bot check.
The company's attention has widened.
Evidence: Since August 2026 the homepage sells inference infrastructure, GPU sandboxes and research partnerships; Studio is one product among several, and Omnia, announced as a base model in February 2026, has no page in the public model catalog.
Uncertain
Whether consent is enough to animate someone else.
Why unsure: The Acceptable Use Policy bans inputs or outputs showing anyone other than yourself, except people who died at least 100 years ago, while the face swap help page only asks for permission. Hedra does not say which rule it enforces.
Whether a free tier still exists after sign-up.
Why unsure: The pricing page shows none, yet its FAQ says a cancelled account keeps limited watermarked generations, which suggests a free level inside the app.
Whether per-model credit costs changed with the new plans.
Why unsure: Hedra published the new plan prices but no credit table, so the true price rise per video can only be judged inside the app.
How Hedra's own models compare with the rivals it hosts.
Why unsure: No public test has run the same portrait and audio through Character 3, Hedra Avatar, HeyGen Photo Avatar 4 and OmniHuman and compared the results.
The numbers, each with its source and the date we checked it.
| Spec | Value | Note and source |
|---|---|---|
| Company | Hedra Inc., San Francisco; founder and CEO Michael Lingelbach | Source: hedra.comchecked |
| Funding | US$32 million Series A in May 2025, led by Andreessen Horowitz's infrastructure fund; valued at US$200 million, per a Reuters source | Source: finance.yahoo.comchecked |
| Own video models | Character 3 (March 2025) and Hedra Avatar | Source: hedra.comchecked |
| Avatar specs | Image plus audio to video; 540p, 720p or 1080p; up to 10 minutes; seven aspect ratios | Source: hedra.comchecked |
| Avatar API rate | 2.5 cents a second at 540p, 5 at 720p, 6.25 at 1080p | Source: hedra.comchecked |
| Hosted models | 110 in the live catalog: 53 video, 40 image, 12 audio, 5 text | Source: api.hedra.comchecked |
| Basic plan | US$20 a month, 2,000 credits, slower generations, commercial use | Source: hedra.comchecked |
| Pro plan | US$50 a month, 7,200 credits, faster generation, commercial use | Source: hedra.comchecked |
| Ultra plan | US$100 a month, 18,000 credits, fastest generation; allowances up to 57,600 credits | Source: hedra.comchecked |
| Teams plan | US$75 a month, 14,400 credits, fastest generation, Teams plan access | Source: hedra.comchecked |
| Plans before late September 2026 | Basic US$15 for 1,500 credits, Creator US$30 for 5,400, Professional US$75 for 14,400 | Source: web.archive.orgchecked |
| Credit rollover | Plan credits reset each month; credit-pack credits never expire but need an active subscription | Source: hedra.comchecked |
| Annual plans | Billed for the year at once, credits granted month by month | Source: hedra.comchecked |
| Free use | No free plan on the pricing page; cancelled accounts keep limited watermarked generations | Source: hedra.comchecked |
| Payment | Visa, Mastercard, Amex, debit and prepaid cards through Stripe; no PayPal, UPI or crypto | Source: hedra.comchecked |
| API billing | Prepaid US-dollar wallet starting at US$0; Studio credits do not pay for API calls | Source: hedra.comchecked |
| MCP server | mcp.hedra.com, for Claude, ChatGPT, Cursor, Claude Code, Codex and Hermes | Source: hedra.comchecked |
| Rights in outputs | You own inputs and outputs; Hedra takes a perpetual, irrevocable licence to use them to run the service | Source: hedra.comchecked |
| Biometric data | Face and voice templates kept no longer than three years after your last interaction | Source: hedra.comchecked |
| Security | SOC 2 Type I; 51 subprocessors listed | Source: trust.hedra.comchecked |
Answer first, then why, an example, the limits and what to do next.
Hedra has no free plan on its pricing page as of November 3, 2026, and its model pages say generation on Hedra uses paid credits. The cheapest Studio plan is Basic at US$20 a month for 2,000 credits.
Limits
Until at least October 7, 2026 the pricing page said 'Start free, upgrade anytime', and its FAQ still says a cancelled account keeps limited watermarked generations, so a small free level may remain inside the app.
Hedra Pro costs US$50 a month for 7,200 credits, with faster generation than Basic and commercial use. It replaced the US$30 Creator plan, which gave 5,400 credits, in late September 2026.
Why
Per credit, Pro works out about 25% dearer than Creator was, by our arithmetic. Ultra, at US$100 for 18,000 credits, adds the fastest queue and Teams access.
Limits
The public pages do not say how many credits a minute of Character 3 video costs, so check the live selector after signing in.
Hedra raised its Studio prices between September 14 and September 29, 2026, per archived copies of its pricing page. Basic went from US$15 to US$20, Creator at US$30 became Pro at US$50, and Professional at US$75 became Ultra at US$100, each with more credits.
Limits
Teams stayed at US$75 for 14,400 credits, now a lower price per credit than Ultra. Annual plans bought before the change keep their full-year credit grant.
Hedra's monthly plan credits do not roll over; they reset to the plan's amount each cycle, and annual plans also grant credits one month at a time. Credits from one-time credit packs never expire.
Limits
Pack credits can only be spent while you hold an active subscription, so cancelling freezes them until you resubscribe.
Hedra lists commercial use on every paid plan, from Basic at US$20 a month upward, and its terms say you own your inputs and outputs. You still grant Hedra a perpetual, irrevocable licence to use your content to run the service.
Limits
Owning the output does not clear the people, voices, brands or music in it, and the Acceptable Use Policy bans media of anyone other than yourself.
Hedra is an established company: Hedra Inc. of San Francisco raised a US$32 million Series A led by Andreessen Horowitz in May 2025, as Reuters and SiliconANGLE reported, and its trust center lists a SOC 2 Type I report. Its status page showed no website incidents from August to November 2026.
Why
The risks are in the details: the terms send disputes to binding arbitration with a class-action waiver, the content licence is broad, no refund policy is published, and no app in the US App Store comes from Hedra Inc.
Hedra's Acceptable Use Policy says no: it bans inputs or outputs containing photos, audio or video of anyone other than yourself, except people who died at least 100 years ago, and bans deepfakes that impersonate real people. Its face swap and voice cloning pages add that you may only use faces and voices you have permission to use.
Limits
The two rules differ, and Hedra does not say which one it enforces; written consent is the minimum before animating anyone else.
Hedra's privacy policy allows it to use personal information, including data derived from faces and voices, to improve the service and develop new products, and to train its AI models on de-identified data. The policy lists no opt-out from that use.
Limits
Face and voice templates are kept up to three years after your last interaction, deleted data stays in backups for up to 30 days, and prompts and media go to the outside model providers you pick.
Hedra has a REST API, version 3, with Python and TypeScript SDKs, a CLI and about 110 models behind one key. It bills a separate prepaid dollar wallet, and Studio subscription credits cannot pay for API calls.
Example
Hedra Avatar costs 6.25 cents a second at 1080p on the API, so a one-minute clip is US$3.75.
Limits
The exception is the MCP server, where jobs can draw on a Studio subscription. Rate limits depend on your key and account and are not published as fixed numbers.
Hedra Character 3 is Hedra's own model that animates one portrait from an audio track into a talking or singing video, with lip sync, facial expression and gestures. Hedra dates it to March 2025; it renders up to 10 minutes at 540p to 1080p.
Limits
Hedra's own model page notes that users report over-dramatic hand gestures, which a prompt can restrain, and its newer Hedra Avatar model carries the same specs and price.
Choose HeyGen for business presenter videos, translation and a free tier of three watermarked videos a month; choose Hedra for expressive characters of any style, a creative agent and access to many other makers' models on one plan. HeyGen Creator costs US$29 a month and Hedra Basic US$20 on November 3, 2026.
Why
Hedra even hosts HeyGen's Photo Avatar 4 model on its API, at 10 cents a second against 6.25 for its own Hedra Avatar at 1080p.
Next step
See the [HeyGen alternatives list](/best-heygen-alternatives) for more options.
Hedra's public pages mention watermarks only for cancelled accounts, which keep limited watermarked generations, per the pricing FAQ. The paid plans list commercial use and say nothing of a watermark.
Limits
Hedra does not publish free-tier watermark rules, since its pricing page lists no free plan.
The short version of everything above.
Pros
Cons
This review is built from what could be checked on November 3, 2026: Hedra's own site, pricing page, model pages, live API catalog, help center, developer docs, blog, terms, privacy and acceptable-use policies and trust center; archived copies of its homepage and pricing page in the Wayback Machine; reporting by Reuters and SiliconANGLE; a lip-sync test by Tom's Guide; a review by LemonSlice, which sells a rival avatar product; and the pricing pages of six alternatives. Hedra's statements about itself are labelled as its own, and every price is the one shown on that day.
Hedra Inc. is a San Francisco company founded and run by Michael Lingelbach, its CEO. It became known for Character, a family of models that make a still portrait talk or sing in time with an audio track. In May 2025 it raised a US$32 million Series A led by Andreessen Horowitz's infrastructure fund, with Index Ventures, Abstract and a16z speedrun joining, as Reuters and SiliconANGLE reported; Reuters' source put the valuation at US$200 million. At the time Hedra had 20 employees and launched Character-3 with Hedra Studio, its web app. Its about page lists Andreessen Horowitz, Index Ventures, Abstract Ventures and the Alexa Fund as backers.
Hedra sells three things today. Hedra Studio, now pitched as the Creative Agent, is where creators and marketing teams make video, images and audio; you brief an agent in a Space and it picks the models, per the Studio help center. The developer platform puts Hedra's own models and those of other makers behind one API. And, according to its homepage and llms.txt, Hedra is "an AI research lab" that sells inference infrastructure, GPU sandboxes and post-training to companies building their own visual and robotics models. Its enterprise page claims that 20% of the Fortune 500 build with Hedra; that is the company's own figure.
The changes that matter to a Studio buyer, in order:
No news report of the price change or the change of focus turned up in searches; both are documented here from Hedra's own pages and their archived copies.
Hedra makes two of the models it sells. Character 3 and Hedra Avatar take a start image and an audio track, render 540p, 720p or 1080p in seven aspect ratios and run for up to 10 minutes, with the length set by the audio. They carry identical descriptions and identical prices in the catalog; the Avatar page says it runs on Together AI infrastructure. Hedra dates Character 3 to March 2025, and its own FAQ admits users report over-dramatic hand gestures that a prompt can restrain.
Everything else is hosted. Hedra's live API catalog listed 110 models on November 3: 53 for video, 40 for image, 12 for audio and 5 text models. They include Kling V3, Veo 3.1, Seedance 2.5, MiniMax H3, Wan 3.0, Nano Banana, GPT Image 2 and ElevenLabs voices, and also avatar models from rivals. The trust center names the providers behind them, from fal.ai and BytePlus to ElevenLabs, MiniMax, Kling AI and HeyGen.
On the API, talking-head video costs:
| Model | 1080p rate | One-minute clip |
|---|---|---|
| Hedra Avatar or Character 3 (Hedra's own) | 6.25 cents a second (2.5 at 540p) | US$3.75 |
| Kling AI Avatar v2 Standard | 5.62 cents a second, 720p only | US$3.37 at 720p |
| HeyGen Photo Avatar 4 | 10 cents a second | US$6.00 |
| OmniHuman 1.5 | 16 cents a second | US$9.60 |
Rates from the catalog; the per-minute figures are our arithmetic, and Hedra's own HeyGen comparison page shows the same US$9.60 for a 60-second OmniHuman clip. Hedra's own models are the cheapest 1080p avatar option in its catalog.
| Plan | Before late September 2026 | From late September 2026 | Notes |
|---|---|---|---|
| Basic | US$15 for 1,500 credits | US$20 for 2,000 credits | Slower generations, commercial use |
| Creator, now Pro | US$30 for 5,400 credits | US$50 for 7,200 credits | Faster generation |
| Professional, now Ultra | US$75 for 14,400 credits | US$100 for 18,000 credits | Fastest generation; allowances up to 57,600 credits |
| Teams | US$75 for 14,400 credits | US$75 for 14,400 credits | Fastest generation, Teams plan access |
| Enterprise | Custom | Custom | SSO, private deployments, Slack support |
Sources: the current pricing page and the archived copies of September 14 and October 7, all billed monthly.
Pro, which the pricing page labels "Popular", got the biggest rise. By our arithmetic a credit now costs about 0.69 cents on Pro, against about 0.56 on Creator, roughly 25% more; Basic stayed at 1 cent a credit, and Ultra costs about 0.56 cents. Teams did not change, so it now buys credits more cheaply than Ultra. Whether the number of credits per video also changed cannot be checked from outside: the help center says credit prices per model are deliberately left out and shown only in the app's live selector.
The credit rules cut against light users. Monthly plan credits reset each cycle, and since the change annual plans grant credits one month at a time, with unused ones expiring at the next grant, per the annual plans page. Annual plans bought earlier, such as Creator and Professional, still get their full year up front. Credit packs never expire, but the credits FAQ says you need an active subscription to spend them. Payment is by card through Stripe; PayPal, UPI and crypto are not accepted (payment methods).
The pricing page carried the title "Free Plan, No Credit Card" until at least October 7. The current page lists no free plan, and Hedra's model pages, such as the one for Kling, say generation on Hedra uses paid credits. The pricing FAQ says a cancelled account loses most features apart from limited watermarked generations, which suggests a small free level inside the app, but Hedra does not publish its limits.
No refund policy appears on hedra.com. The generation history help says only that a failed request may be refunded automatically, and the terms renew subscriptions automatically until you cancel.
The developer platform is the most open part of Hedra. Per its October 6 post, one key reaches every catalog model through typed asynchronous jobs; an estimate call prices the exact request for free; webhooks are retried for about six hours; and Python and TypeScript SDKs, a CLI and OpenAI- and Anthropic-compatible endpoints for the text models are available. Its quickstart prices a six-second MiniMax H3 clip at 768p at US$0.36.
Billing is separate from Studio. The API draws on a prepaid dollar wallet that starts at US$0, and Studio subscription credits cannot fund it (API billing). Rate limits depend on the key and account rather than a published quota (errors and limits). Hedra's own comparison of one 10-second, 1080p job with sound runs from US$0.60 on LTX-2.3 Fast to US$7.91 on Seedance 2.5.
The MCP server at mcp.hedra.com works with Claude, ChatGPT, Cursor, Claude Code, Codex and Hermes after a Hedra sign-in, and it is the one place where a Studio subscription can pay for jobs. The terms warn that requests from a connected assistant count as your instructions and can spend credits or trigger an auto-recharge.
An outlet test, on an older model. In October 2024, Ryan Morrison of Tom's Guide ran Hedra's Character-2 against Kling and Runway on 10-second lip-sync clips from three Midjourney portraits. Hedra won the neutral-face round for its voice and mouth movement, won a round with a wide smile showing teeth only because every tool did badly, and lost a side-on, mid-gesture round to Runway; it finished first overall. That test predates Character 3 and Hedra Avatar, and no outlet test of the current models or the Studio agent turned up in searches.
A competitor's review. LemonSlice, which sells real-time interactive avatars, reviewed Hedra from its product pages on September 11, 2026. It found that everything Hedra ships is pre-rendered, with no live avatar product, that the avatar models stop at 1080p and that Hedra publishes no typical render times. Its price table, captured that month, matches the old plans in the archived pricing page.
Hedra's own numbers. In its agent benchmark, removing a planning step raised the agent's success rate on recorded scenarios from roughly 40% to 60%. Hedra designed and ran that measure.
User reports. Hedra's Trustpilot page answered with a bot check and Reddit's robots.txt disallows crawling, so neither is quoted here. The status page showed all systems operational and no website incidents from August to November 2026.
Hedra's written rules are strict. Its Acceptable Use Policy bans inputs or outputs that contain personal information of anyone other than yourself, including photos, audio and video of other people, with an exception for people who died at least 100 years ago. It also bans deepfakes that impersonate real people or mislead about a clip's origin without consent or legal right, sexually explicit content, election interference and uploads of sensitive personal data.
The how-to pages are looser. The face swap help says to use only faces you have permission to use, and the voice cloning page says to clone only a voice you have permission to clone, without the AUP's stricter limit to your own face and voice. Hedra lists a content-moderation provider among its subprocessors but does not describe how it screens uploads. Anyone making ads with an actor, a client or a staff member should get written consent and expect that the AUP may still forbid it.
The privacy policy, effective September 4, 2026, says Hedra may derive biometric data from faces and voices to provide its avatar features, and may use personal information, including that derived data, to improve the service and develop new products. It may also train its AI models on aggregated or de-identified data. The policy's list of your choices includes no opt-out from that use.
The biometric data policy promises written consent where Illinois, Texas or Washington law requires it and caps retention of face and voice templates at three years after your last interaction. Deleted data can stay in backups for up to 30 days. The terms leave you the owner of your inputs and outputs but give Hedra a perpetual, irrevocable, sublicensable licence to use them to run the service for you and other users. Disputes go to binding arbitration with a class-action waiver, with a 30-day window to opt out. The trust center lists a SOC 2 Type I report and 51 subprocessors. A search of the US App Store on November 3 found no app published by Hedra Inc.; an app with Hedra in its name comes from an independent developer.
| Option | What it is | Entry price on November 3, 2026 |
|---|---|---|
| Hedra | Its own talking-head models, a creative agent and about 110 hosted models | Basic US$20 a month for 2,000 credits; API pay as you go |
| HeyGen | Presenter avatars, video agent and translation | Free: 3 watermarked videos a month up to 1 minute; Creator US$29 a month, 600 credits |
| Synthesia | Business presenter avatars and dubbing | Free Basic with a watermark; Starter US$19 a month, or US$14 billed yearly |
| Captions | Phone-first editor with AI actors | Max US$24.99 a month, 500 credits (iOS price) |
| Magic Hour | Lip sync, face swap, talking photos and 30-plus models | Creator US$19 a month, or US$12 billed yearly; credits never expire |
| Runway | Its own Gen-4.5 plus hosted models | 125 one-time free credits; Standard US$15 a month (US$12 yearly) |
| Higgsfield | Many makers' video and image models | Starter US$19 a month billed yearly, 270 credits |
Sources: HeyGen, Synthesia, Captions, Magic Hour, Runway and Higgsfield pricing pages.
HeyGen is the better choice for realistic business presenters, translation and a free tier to start, and its monthly credits roll over for one month where Hedra's reset. Synthesia suits training and corporate teams that want dubbing, guest seats and a free allowance. Captions fits creators who edit on a phone. Magic Hour is the pick for occasional use, because its credits never expire and its face swap and lip sync sit next to many hosted models. Runway and Higgsfield are stronger for cinematic video generation than for talking heads. Hedra keeps the edge for expressive characters of any style, long talking-head clips at a low API rate, and one account across avatar, video, image and voice models. More options are in the HeyGen alternatives list.
Sources
Changelog
Riverside is a browser-based studio from RiversideFM, Inc. that records podcasts and video interviews locally in up to 4K, then transcribes, edits and clips them with AI. On October 26, 2026 it added four AI agents to every plan. Plans cost $35 a month or $24 to $29 billed yearly; its terms let Riverside train AI on your recordings.
Luma AI is the creative-agent app and API of Luma AI, Inc., which replaced Dream Machine with Luma Agents in March 2026 and runs its Ray3.2 video model beside Kling 4.0 and Seedance 2.5. Plans start at US$30 a month for 10,000 credits, with no free plan listed. Ray3.2 sits 33rd of 34 on Artificial Analysis's silent image-to-video board.
OpenArt is a web studio by San Francisco's OpenArt, Inc. that runs 100+ image, video and audio models from ByteDance, Google, OpenAI and others under one credit balance, plus Director, which turns a chat into a video of up to five minutes. Plans cost US$14 to US$240 a month, commercial use from US$34. Credits run out fast and are non-refundable.
Magic Hour is a browser studio and API from Y Combinator-backed Magic Hour AI, Inc. that pairs its own face swap and lip sync tools with 30-plus hosted models such as Kling and Veo 3.1. Kling 4.0 arrived on October 27, 2026. Plans start at US$19 a month for 12,000 never-expiring credits; premium models use them fast.
Runway is an AI video platform from New York's Runway AI, Inc. that runs its own Gen-4.5 and Aleph 2.0 models beside Kling 4.0, Veo 3.1 and Seedance on one credit balance. Individual plans cost US$15 to US$95 a month; the legacy Unlimited plan ends November 30, 2026. Gen-4.5 now ranks 28th on Artificial Analysis's silent text-to-video board.
Pika H3 Flex is Pika's six-step, post-trained version of MiniMax H3 that makes 5- to 15-second clips with stereo sound, released with open adapter weights on October 29, 2026. Pika's API charges 0.04 USD a second at 768p. The weights' licence excludes the US, EU, UK and South Korea, and Pika's package is qualified only on 80 GB H100s.