Privacy Policy

1. Translation: captions first, then audio

When you click “Translate video”, VidClara first tries the local report for the same video and target language. If there is no matching valid result, it reads accessible captions for the selected video. Existing captions send only their text, timing, language and necessary metadata over HTTPS to VidClara, then Google for translation. Without captions, the extension attempts to upload the complete selected video. The default translation route extracts and validates its full audio track, converts it to 16 kHz mono PCM, losslessly compresses it as FLAC and sends that audio to Google for transcription. A second request sends the transcript text for translation. This route does not generate visual analysis or send video frames to Google.

If the complete file cannot be read, is too large or has only a temporary streaming address, VidClara automatically tries to record audio from only the selected HTML video player. It starts that video from the beginning at normal speed and attempts to restore its previous position, speed, loop and play/pause state after completion, cancellation or failure. It preserves mute and volume. It does not use the microphone or capture the screen, whole tab or other players. If the media has changed, it does not restore the old state onto the new video. Audio remains temporary in browser memory and is not automatically saved as a file.

Each video is limited to 180 seconds. Complete-file uploads are limited to 12 MiB, audio fallback to 2 MiB, and local files to 12 MiB. Capture needs approximately the video’s playback time before cloud processing. Cross-origin, playback or content-protection restrictions can prevent access; VidClara does not bypass these restrictions.

2. Full visual analysis and optional product details

When you click “Analyze video” or “Analyze scenes and script”, VidClara sends the actual complete video to its service and Google. This task can produce original speech, a target-language translation, on-screen text, a shot timeline, sound observations and script-structure analysis. A transcript or audio track is not substituted for visual input. Silent videos can be analyzed visually. The model can miss brief frames, text or speech; its timecodes are not a claim of frame-perfect measurement.

If the original file is unreadable, VidClara automatically tries to record video and available audio from only the selected player’s media stream, playing from the beginning at normal speed. It attempts the same state restoration described above. It does not capture your microphone, screen, whole tab or other players. This video capture is limited to 12 MiB and an actual duration of 180 seconds; site restrictions can still block it.

If you choose to enter your own product facts, target market or production requirements, those details are sent with the video to VidClara and Google to generate a corresponding rewrite. Their combined UTF-8 limit is 4,096 bytes. Without your product facts, the task analyzes the reference video rather than treating the reference product’s claims as facts about your product. Shot prompts are suggestions for a new creation, not recovery of the original creator’s prompts. Do not enter API keys or unrelated sensitive information. Product context helps distinguish cached tasks; reports and current task state can retain related generated content.

3. Consent, providers and model requests

Before first cloud use, a one-time dialog in the extension explains where captions, audio, video and optional details are processed. No media, captions or product details are uploaded before agreement. Cancelling starts no task. Agreeing resumes only the selected task and saves consent locally for subsequent one-click use. Service and allowance checks may establish an anonymous session first, without media, captions or product details. The website stores only a manually chosen page language in your browser and does not invoke a translation model.

Several windows can wait for the same cloud task. Each invocation uses a random temporary subscription ID; its waiting association is held only with the task in server memory, not as a new persistent installation identifier. Cancelling leaves only that subscription; the service attempts to stop shared work only after the last subscription leaves. A polling/network interruption does not automatically cancel an accepted task. An ordinary retry tries to reuse it, while explicitly choosing “Skip cache” creates a new task under the normal quota rules. Cancellation before the client receives a job ID may not immediately locate accepted work; server deadlines and quota limits still apply. Cancellation or failure cannot guarantee withdrawal of Google processing or a provider refund.

Alibaba Cloud supplies infrastructure and network services; Google supplies model processing. Google’s data use and retention depend on the project, service configuration and applicable terms. VidClara does not promise zero Google retention or that every service tier excludes product improvement. See the Gemini API terms and Google Privacy Policy. Submit only material you are entitled to process and do not submit private material you do not want processed in the cloud.

Full visual analysis normally uses one model request. Audio translation with speech normally uses two: transcription, then text translation; no-speech audio skips translation. Existing captions normally use a text-translation request. Temporary connection, timeout or some server failures can trigger bounded retries with backup keys. The audio stages share a maximum of five actual requests and a 120-second task deadline; each stage can try at most four distinct keys. Rate-limit responses trigger a pool cooldown, not quota evasion. Failed or timed-out provider requests can still incur operator costs; switching keys does not consume another user task.

4. Server storage and retention

5. Anonymous sessions and shared free limits

No VidClara account is required. A random installation identifier is generated and kept in extension-private storage. Signed service sessions are bound to the source IP and refreshed after expiry or a network change. The service uses HMAC-hashed installation and IP dimensions for daily limits; neither raw IPs nor publisher API keys are returned to video pages.

Translation and full visual analysis share up to 50 tasks per installation per UTC day, subject to shared limits of 250 per IP and 1,000 globally. A new task reserves one allowance. A failure or cancellation is refunded only when the service can establish that no provider request was attempted; otherwise it remains counted. Separate preparation-request abuse counters are not refunded. Valid cache hits and reuse of the same server job do not create a new task. Counts measure user tasks, not provider request counts or fees. Shared limits and availability mean 50 tasks are not guaranteed to every installation. These protections are not paid unlocks.

6. Browser storage, language and deletion

Local storage holds interface/target-language and caption preferences, the random installation identifier, anonymous session token and cloud-consent record. Up to 10 recent reports within about 4 MiB can remain locally. Reports can contain original/translated speech, screen text, scene and script analysis, product-related generated content and a filename; they do not contain video files, source URLs or publisher API keys. Selected media and current task state are stored in the extension’s session area.

Clear saved captions in Settings, or remove the extension and its data to clear local reports, preferences and consent. Deleting local data cannot recall content already sent to Google or remove files you exported separately. Server results expire as described above.

Interface language follows the browser or your manual choice, without IP geolocation. Interface dictionaries are bundled locally and are not translated by AI. Target language is sent with the requested model task. Changing interface language does not start a model task. Opening the sidebar synchronizes the same video’s existing result without a new task. Private session matching retains up to 12 candidate associations for approximately 30 minutes, expiring during later updates. Window-matching identifiers are not sent to the model. Closing the sidebar does not automatically stop a background task.

Copy and export occur only after your action. SRT, Markdown, CSV and JSON files are saved through your browser; you manage their deletion and sharing.

7. Video downloads and cover previews

YouTube video-file downloads are not provided. YouTube translation, analysis and subtitle exports remain available. Other sites support only readable complete media that you are entitled to download.

When you request a download or open the Downloads workspace, VidClara reads media addresses, titles, page addresses, available durations and thumbnails for videos already loaded on the supported page. While the workspace is visible, it scans approximately every two seconds, retaining up to 50 deduplicated candidates. It does not automatically scroll or crawl an account. Candidates expire for reuse after 10 minutes without an update; expiry does not mean immediate erasure of stored state.

Candidates, selections, queue state, thumbnails and internal media addresses stay in extension-private session storage. Sidebar snapshots omit media and original cover URLs. Cover collections for candidates, queues and recent individual downloads are each bounded to about 512 KiB; recent individual covers retain at most 10 items. An in-memory remote-image cache holds at most 20 items within about 512 KiB and clears when the background process closes.

Downloads go directly from the media site to your browser, without VidClara cloud, Google or a model call. Browser downloads can use credentials under browser/site rules. Previews first use a local frame or a same-video card image, without seeking the player. Only hover or keyboard focus, when no local image is available, can request a small JPEG, PNG or WebP from a permitted image site, without browser credentials or redirects. These media/image sites receive the corresponding requests.

Closing the sidebar does not cancel browser transfers or remove downloaded files or download history. Cancel through the extension or browser downloads list, and manage saved files/history yourself. VidClara supports readable complete files and does not bypass login, payment, DRM or arbitrary stream restrictions. Download only content you are entitled to save.

8. Permissions, diagnostics and this website

Declared website permissions find the selected video, read accessible captions/media and display captions or control playback. Video pages receive only that video’s results, display preferences and playback-control messages; they cannot read extension-private storage. Model output is displayed as text and never executed as instructions or HTML. The extension does not directly read browser cookies, account storage or the clipboard. Reading captions/media still makes requests to their source websites.

The public package excludes direct-to-Google and Native Messaging permissions and contains no publisher model keys. The development preview’s optional local helper and personal API connections are separate advanced features; this policy describes the public service workflow. This release has no VidClara cloud account sync or payment collection.

“Copy diagnostics” copies only the product/version, service-configured flag, processing route, interface/target languages, workspace and recent status after your click. It excludes video, captions, filenames, URLs, keys, tokens, installation IDs and raw error bodies. Nothing is sent automatically. Feedback opens your mail app; you choose what to paste and send.

This product website includes no analytics scripts, advertising trackers or collection forms. Hosting infrastructure may keep operational logs under its own rules. External links and email links use their respective services.

9. Purpose, Limited Use and changes

VidClara’s use of user data complies with the Chrome Web Store User Data Policy, including Limited Use restrictions. Data is used only for requested translation, visual/script analysis, product-related generation, result display, downloads and necessary security/operations. We do not sell user data or use it for advertising profiles or credit assessment. We do not manually read content except with explicit consent for support, as necessary for security, or as required by law. Necessary providers process data for the disclosed purposes.

We will update this page and its date when processing practices change. AI recognition, translation, visual analysis and generated rewrites can contain errors; check important details against the original video and your own product facts.