PlayHT has shut down
If you are here because a link stopped working or an API key stopped authenticating, that is not your configuration. The platform is gone. Here is what happened, what you can still get back, and how to rebuild — including the options that are not us.
What happened
Meta acquired the PlayAI team in July 2025. The service was wound down afterwards and closed at the end of 2025; the shutdown notice on their own home page read simply “We have shut down the service. Thank you for being part of our journey.” According to the migration guide the successor vendor published, accounts, saved audio and voice clones were deleted at sunset. play.ht and play.ai no longer resolve — the domains return no address record at all. (checked 2026-09-03).
We do not publish an exact closing date, because PlayHT never did. The figure that circulates traces to secondary write-ups; what can be checked is that their site was still an ordinary marketing page on 31 December 2025 and carried the shutdown banner by 6 January 2026.
- techcrunch.com — read 2026-09-03
- web.archive.org — read 2026-09-03
- inworld.ai — read 2026-09-02
What you can still recover, and what you cannot
Recoverable
- Audio you downloaded.Files on your own disk, in a client’s project folder, or already published are unaffected.
- The original recordings you cloned from. If you kept the source samples — the raw voice recording, not the synthetic output — you can clone again anywhere.
- Your scripts. Text is the cheap half of this. Anything still in a document is a re-generation away.
Not recoverable
- The clone itself.A voice model is not portable between vendors, and PlayHT’s were deleted. No platform can import another’s clone — the weights were never yours to export.
- Audio you only ever streamed or previewed. If the file lived in their account and you never pulled it down, it is gone.
- A clone rebuilt from generated audio. Cloning from a synthetic sample compounds artefacts, and if the voice was not yours you may no longer have a consent record to point at. Go back to the source recording.
How to rebuild, in order
- 01Find the original recordings. Not the PlayHT output — the human recording you cloned from. A clean minute of speech with no music under it is worth more than ten noisy ones.
- 02Check your permission is still in order. If the voice is not your own, you need the speaker's explicit permission, and it is worth having it in writing before you clone anywhere. Our Responsible Use policy sets out what we require, and the audio rights page sets out what you may then do with the result.
- 03Re-clone, then re-cast what you cannot clone. Your custom voices come back from the samples. Stock voices do not: PlayHT's catalogue is not ours, so anything narrated by one of their presets needs a new voice chosen. Do this before you re-generate a library, not after.
- 04Re-generate from your scripts. Then download everything, and keep the files. The lesson of this shutdown is not which vendor to pick — it is that the copy on your own disk is the only one nobody can switch off.
- 05Repoint your integrations. If code was calling their API, it now calls ours at /v1 with an x-api-key header, or reaches the studio over MCP if the caller is an AI agent. The API reference documents every operation.
What maps to what
| DIMENSION | VoiceLabs | PlayHT, until it closed |
|---|---|---|
| Instant voice clone | Instant cloning from a short reference clip, planned at launch. You will need the original recording again — a clone cannot be reconstructed from generated audio, on any platform. | A cloned voice built from a short sample, stored in your PlayHT account. Deleted at sunset along with the account. |
| Preset voice library | Preset voices from seven TTS engines — Qwen3-TTS, Chatterbox, Kokoro and others — chosen per generation, with 23 languages across them. Different voices, not a like-for-like replacement: you will be re-casting, not re-mapping. | Their own catalogue of stock voices, selected inside their studio. |
| REST API | A REST API at app.voicelabs.now/v1 with x-api-key authentication, RFC 9457 error documents, per-key scopes, rate-limit headers and an Idempotency-Key on writes. API access is part of VoiceLabs Pro. | An HTTP API with an API key, retired when the platform closed. |
| Generated audio you had saved | Downloadable from the moment it is generated, on any account. Keeping your own copy is the only migration plan that survives a vendor closing, ours included. | Held in the account, and gone with it. |
| Agent / assistant access | A remote MCP server over streamable HTTP with OAuth 2.1 and a consent screen, revocable per agent — nothing to install and no key to paste. Live today. | API key pasted into whatever tool needed it. |
Where else to go
We are not the only answer and it would be insulting to pretend otherwise to someone who has just lost a workflow.
- Inworldpublished the migration guide PlayHT’s users were pointed at and positioned itself as the successor. If you want the shortest path off a dead API, start there.
- ElevenLabs is the category leader, ships today, and does far more than we plan to — our comparison says as much.
- Fish Audio has the most generous free tier in this market, including cloning and commercial use, which is worth knowing if your budget just became uncertain — our comparison.
- VoiceLabs, once the studio ships: one flat $8 a month for 2 hours of generated audio, seven engines, cloning included, downloads never withheld, and an MIT open-source foundation you can read.
7-day free trial, card upfront, cancel anytime before it ends.
PlayHT, PlayAI, Meta, Inworld, ElevenLabs and Fish Audio are trademarks of their respective owners. VoiceLabs is not affiliated with, endorsed by, or sponsored by any of them, and has no relationship with the PlayHT shutdown beyond writing this page.