Effective 2026-09-30Version 2026-09-30This English text is the governing version. A Chinese translation is provided for convenience; where the two differ, this English text prevails.
1. What Vidleaf is
Vidleaf is a reading workbench for video subtitles. You give it a video link; it
retrieves the subtitles the video platform already publishes. When no subtitles
exist, and only if you ask for it, it sends the audio for transcription. You can
also upload audio you are entitled to process (common audio formats) and have it
transcribed after you confirm. If you choose a video file, the page extracts its audio track
on your device and uploads only the sound; the video picture never leaves your device. It can
also translate, summarise, separate speakers, and let you annotate and search
what you are reading.
Note the boundary this draws: Vidleaf sells processing, not a licence to video
content. We do not own, license, or claim rights in any third-party video.
Whether you may access or use a given video and its subtitles is between you,
the platform, and the rights holder. Paying us does not change that.
2. Using Vidleaf with and without an account
You can use part of Vidleaf without signing in.
No account
Account required
Open a link and read existing subtitles
Yes
Search, switch script, export what you are reading
Yes
Audio transcription
Yes
Translation and AI summaries
Yes
Speaker detection
Yes
Reading history and your own speaker labels
Yes
The line is not arbitrary. Everything on the left costs us nothing per use and
sends nothing to a third party. Everything on the right either costs money per
use or sends audio abroad, and both of those need an account to meter and to
record your consent against.
3. Accounts
Reading a video that has already been processed needs no account. An account
is needed for the operations that cost money — transcription, translation,
summaries, speaker detection — and for keeping a history.
There are three ways to register. Pick one:
Route
What we learn about you as a result
Continue with Google
Only an identifier Google issues for you on this site, and a display name. We do not store your Google email address, and we do not store your picture.
Continue with email
Your email address is stored, both to receive a verification code and to sign in later. We send that address a six-digit code first.
Invitation token
A username, a display name and a password. This route involves no email address.
We do not ask for a phone number, a real name or identity documents on any
of the three routes.
One email address can register one account.
Passwords are at least 8 characters and must not be easy to guess: we refuse
known common passwords, all-digit passwords, sequential or repeated characters,
and passwords that contain your username, your email name or this site's name.
Passwords are stored only as an irreversible
hash. Nobody, including us, can read your password back. An account created
with Google has no password.
Forgetting your password has two cases:
Registered with email: press "Forgot password" on the sign-in page and we
send that address a reset link. It is valid for one hour and can be used once.
Resetting signs you out on every other device.
Registered with Google or an invitation token: no self-service recovery.
A Google account has no password in the first place, and for an invitation
account we hold no email address, so we cannot find it. Ask us in those cases.
One account per person. Accounts may not be transferred, rented or shared.
You are responsible for what happens under your account.
4. What you may use it for
Personal study, research and reference. The specific permitted and prohibited
uses are in the Content Rules, which form part of these Terms.
5. Quotas and automatic pause
Registered accounts may read subtitles that already exist without a reading limit. Free processing uses the following audio and AI allowances. Free speaker detection shares the audio allowance; Pack S and L speaker entitlements are metered separately.
Free account
Audio transcription
30 minutes / week
AI features (summaries, translation)
10 per week
Maximum duration for new audio transcription
120 minutes
Free allowances and purchased entitlements are metered separately. This section and §6 describe the product rules; current availability, prices and first-order eligibility appear on the plans page and in your account.
A newly registered account receives a one-off extra 30 transcription minutes
and 10 AI uses, on all three registration routes. That extra does not expire
and is not cleared by the weekly reset; it lasts until you use it. It raises
your ceiling and does not alter how we meter our own third-party costs.
Base weekly quotas reset every Monday at 00:00 Beijing time (UTC+8) and do not roll
over. Unused quota is not carried into the next week.
AI features are counted per video: one summary of one video counts as one,
regardless of how many model calls it took behind the scenes.
At most one job running and three queued at a time. Submissions beyond that are
refused; wait for the earlier ones.
Free processing pauses when its budget is exhausted. Purchased processing has a separate budget and requires explicit selection and confirmation. A service pause or outage can affect new tasks; saved subtitles remain readable.
These numbers may change. Changes are announced on the site beforehand.
6. Payment and refunds
Pack S, Pack L and Single Pass are one-time purchases with no automatic renewal; prices include applicable taxes. Only products explicitly available on the plans page and in your account can be purchased. Checkout is provided by Waffo and billed in US dollars; RMB payment is not currently offered. Before payment, we show the entitlements, validity, amount payable, component price allocation and refund policy, and require your separate agreement. Each order retains the accepted policy version and displayed language; existing orders keep their original terms.
S, US$4.99: 600 transcription minutes, 120 speaker-detection minutes and 20 AI uses, valid for 12 calendar months from successful payment. The three allowances are independent and are not interchangeable.
L, US$14.99: 3000 transcription minutes, 600 speaker-detection minutes and 100 AI uses, up to 180 minutes per asset, priority queue and 12 calendar months of validity from successful payment. S allows up to 120 minutes per asset and does not include priority queue.
Single pass, US$1.99: one asset up to 120 minutes, with 1 transcription, 1 speaker detection on supported sources and 2 AI uses shared by summaries and translations. Start within 30 days of purchase; remaining uses stay valid for 30 days after the first use. Your account shows the actual deadlines. Unused processing duration cannot be transferred to another asset.
Free allowances, S and Single Pass allow assets up to 120 minutes. Selecting L entitlements allows assets up to 180 minutes; processing must also fit the selected entitlement and remaining allowance. Uploaded audio files remain limited to 60 MiB. AAC audio in an m4a file, and an AAC audio track extracted on your device from an mp4 or mov video, are not subject to the 60 MiB limit if the average bitrate is at most 320 kbit/s; the per-asset duration limit above applies instead (an m4a over 60 MiB is first checked on your device, and only its AAC audio track is uploaded). A local upload supports automatic speaker detection once it has been transcribed, provided its original is still within the retention window described in the Privacy Policy (around six hours after upload); after that, the same file must be uploaded again. Supported sources are explained before purchase and use. Original-file retention follows the Privacy Policy and is separate from pack validity.
New tasks do not use purchased entitlements by default. Select purchased credits or a single pass, review the estimated free and purchased use, and confirm before submission. Exhausting the free processing budget never automatically spends purchased credits. Expired entitlements cannot start new processing. We do not promise automatic refunds at expiry.
First-order offer: when the offer is active and an account has no paid orders of any product, its first S or L purchase may use the displayed first-order price, once per account: regular S US$4.99 becomes US$2.49 and regular L US$14.99 becomes US$7.49. Prices are 50% of the regular price, rounded down to cents, and cannot be combined with other offers. Single Pass is excluded from the discount, but paying for it also ends first-order eligibility; a refund does not restore eligibility. An unpaid discounted order reserves eligibility until checkout expires and the payment channel confirms it was not paid; an unknown or processing result keeps the reservation. Eligibility and price are checked again before creating an order; existing orders retain their agreed price.
Ordinary refund requests must be submitted within 14 days of successful payment. An unused order is refunded in full. For a used order, the refund is based on the unused share of each component at the prices allocated before purchase, never exceeding the amount actually paid:
Product
Transcription allocation
Speaker allocation
AI allocation
S
US$2.49 / 600 minutes
US$1.00 / 120 minutes
US$1.50 / 20 uses
S first order
US$1.24 / 600 minutes
US$0.50 / 120 minutes
US$0.75 / 20 uses
L
US$7.48 / 3000 minutes
US$3.00 / 600 minutes
US$4.51 / 100 uses
L first order
US$3.74 / 3000 minutes
US$1.50 / 600 minutes
US$2.25 / 100 uses
Single pass
US$0.99 / 1 use
US$0.40 / 1 use
US$0.60 / 2 uses
The refund equals the sum, across components, of allocated price × unused entitlement ÷ original entitlement. S and L transcription and speakers use metered seconds; AI uses task counts. Single-pass transcription and speakers each use one eligibility count, not remaining audio seconds. The cash total is rounded once to the nearest cent, with half cents rounded up; individual components and chunks are not rounded separately.
Review and confirm the refund quote and remaining components before submitting. A change in use requires a new quote. Refunds use the entitlements delivered at the time of the request. Reserved work not yet delivered is treated as unused and frozen; later outcomes are reconciled separately, without waiting for every running task to finish before accepting a request.
An order can have only one successful refund. A successful refund closes all its remaining entitlements. You cannot obtain a small refund and continue using the rest of that order. If a result is unconfirmed, check the original request rather than submitting a new one. The time for money to arrive depends on the payment channel.
Ordinary self-service refunds are unavailable after 14 days, and there is no automatic settlement of unused entitlements at expiry. Non-delivery, duplicate charges and other after-sales issues can still be raised with support@vidleaf.app. This policy does not exclude consumer rights that cannot lawfully be excluded. We use the order record, verified payment-channel records and the policy accepted for that order to resolve requests.
7. Where your data goes
To do what you ask, Vidleaf must send content to third-party providers:
Audio goes to OpenRouter (transcription; it routes the audio to a model provider,
currently DeepInfra, Together AI or Groq) and pyannoteAI (speaker detection). Audio leaves mainland China.
Subtitle text goes to DeepSeek (summaries, translation).
Requests carry no account name and no internal user identifier.
The full list, what each receives, and how long it is kept are in the Privacy
Policy. If you do not want a particular piece of audio to leave this site, do
not use transcription or speaker detection on it.
8. How we obtain Bilibili subtitles — please read this section
Bilibili's subtitle interface returns an empty list to anonymous requests. So that
you can use the site without signing in to Bilibili yourself, Vidleaf holds
Bilibili accounts belonging to its operator on the server and reads subtitle
lists on your behalf using them.
What you should know:
These accounts only read. They do not post, comment, like, follow, or
perform any write action on your behalf.
Each account has a daily call limit and a minimum interval between calls, and
goes into cooldown automatically when the platform rate-limits it.
This is a grey area under Bilibili's user agreement, and the risk is not
hypothetical. In January and July 2026 Bilibili sent infringement notices to
two public projects; the July notice expressly demanded that they stop
scraping and parsing Bilibili content. Our technical approach is of the same
kind. Therefore:
- These accounts may become unavailable at any time, whether restricted by the
platform or withdrawn by us.
- When that happens: Bilibili videos already processed on Vidleaf stay
readable and are unaffected. Videos not processed before cannot be
processed at all, whether or not they have subtitles. Bilibili's subtitle
interface returns an empty list when not signed in and gives no signal
distinguishing "this video has no subtitles" from "you need to sign in" — so
we can neither read the subtitles nor honestly tell you whether any exist.
The site says the collection channel is unavailable; it does not let you
believe the video has no subtitles.
- If you enabled "transcribe when no subtitles are available", we transcribe
directly and it consumes your transcription quota. Note the trade-off: we
were unable to confirm whether free subtitles existed, so that
transcription may have been avoidable. The job log says so.
- We write this here rather than explaining it when it happens. If this risk is
unacceptable to you, do not depend on Vidleaf for Bilibili content.
If you would rather use your own Bilibili session than our shared accounts,
contact us. We will consider offering that. It is not available today.
9. Availability
Vidleaf is provided as is, with no availability commitment and no SLA. It may
be interrupted, may slow down, features may change or be withdrawn, and the
service may stop entirely. If we decide to shut down, we will announce it on the
site in advance and leave time for you to export your own content.
10. Your content and machine output
The links you submit, the labels you write and your search history are yours,
and other users cannot see them.
Subtitles, translations and summaries are cached site-wide per video, not per
account. Another user opening the same video reads the result that already
exists, and you cannot unilaterally delete it. See Privacy Policy §3.
Transcription, translation, summaries and speaker separation are machine
output and will contain errors. The error rate depends mainly on overall
audio quality, background noise and music, people speaking over each other,
accent and speaking rate, and the density of proper nouns and jargon. We do not
control any of these.
You acknowledge and agree that before taking or not taking any action based
on any output of this site, you are responsible for verifying its accuracy and
completeness yourself. It is a reading aid, not a record of fact. Do not use it
as evidence or as the basis of an important decision. Check the original video
for anything that matters.
Speaker labels are marked either certain or inferred. "Inferred" means
exactly what it says: we are not sure who said that line.
11. Labelling of AI-generated content
Content
Produced by
Shown in the interface
Marked in exported files
AI summary
Vidleaf's AI
"AI-generated" at the start of the summary
Yes
AI translation
Vidleaf's AI
"AI translation" when viewing the translation
Yes
Speech transcription
Vidleaf's machine transcription
"Transcription" as the caption source
Yes
Platform human subtitles
The video platform, not us
"Human captions"
No
Platform auto subtitles
The video platform, not us
"Auto captions"
No
Why transcription is not labelled "AI-generated":
transcription writes down what a person in the video actually said; it is not a
machine writing prose in a person's place. Summaries and translations are machine
written text a reader could take for human writing; transcription is not. Marking
a faithful transcript as "AI-generated" would not be safer, only less accurate.
Marking in exported files is written into the file header and carries the
content attribute, the provider name and a content number. The content number is
derived from the video identifier and a hash of the source text, so exporting the
same content twice yields the same number.
One thing we cannot do, stated plainly: the SRT subtitle format has no
metadata layer — the specification has neither a comment syntax nor a header
area. Forcing one in would make players display it as a subtitle line, which
pollutes the content rather than marking it. So SRT exports carry no marking
block; TXT exports and Markdown notes do. Interface labels are unaffected.
Do not tamper with markings, forge them, hide them, or provide tools designed
for that purpose.
12. Complaints and takedown
If you are a rights holder, or represent a video platform, and believe subtitles
or translations held by Vidleaf infringe your rights, email
copyright@vidleaf.app (kept separate from support so it does not get
buried) with:
the video link, and where the content sits on Vidleaf;
the work you claim rights in;
your identity, contact details and basis of your right;
a statement that you believe in good faith the use is unauthorised;
a statement that the notice is accurate and you are authorised to act for the
rights holder.
On receipt we hard-delete that video's subtitles and translations and blacklist
the video, after which we will not process it again. We do not require an
elaborate procedure, and we will not demand that you prove the point beyond
argument — deleting is simply the cheaper option.
13. Violations and termination
Where the Content Rules are breached we may disable your account, delete the
related data, and blacklist the videos involved, without prior notice. You may
stop using Vidleaf at any time; to close an account, email support@vidleaf.app.
14. Changes to these Terms
Changes update the effective date on this page and are announced on the site.
Material changes — lower quotas, a different payment model, wider use of your data
— are announced at least 7 days before they take effect. Continuing to use
Vidleaf does not by itself constitute consent to an updated version. Before starting new cloud operations, signed-in users must explicitly accept the current Terms; declining leaves saved content readable.
15. Disclaimers and limitation of liability
To the maximum extent permitted by law:
Vidleaf is provided "as is" and "as available", without warranty of any
kind, express or implied, including merchantability, fitness for a particular
purpose, accuracy, and uninterrupted operation.
We are not liable for indirect loss, loss of data, loss of profit, or errors in
machine output arising from use of or inability to use the service.
Our total liability to you is capped at the amount you actually paid us in
the 12 months before the claim arose; for users who have not paid, that cap
is zero.
16. Language
The English version of these Terms governs. The Chinese interface is provided
for convenience and is not a separate agreement.