Local testing draft — not finalized for public launch. Last draft update: October 4, 2026.
Audio Studio is the working product name. These draft terms describe the current testing service. They are not a finalized agreement for a public commercial launch. Operator details, contact information, commercial policies, and applicable legal arrangements must be completed before launch. This update date is not a public launch effective date.
Audio Studio is being developed for vocal removal, audio stem separation, and removing background music from video. The current local implementation separates uploaded audio into vocals, drums, bass, and other tracks using Demucs through Replicate. It can also mix the three non-vocal tracks into an instrumental file in your browser.
For video music removal, the browser extracts audio locally and submits that audio through the same separation workflow. It then combines the separated vocal track with the original picture to create a local video download. The original video and exported video are never uploaded by this workflow. Selecting a file only previews it locally; processing begins when you explicitly start it.
Server processing is disabled by default and can only be enabled for local development testing. The current production configuration rejects audio processing. Available features and limits will be shown in the workspace as the product develops.
Sign-in is required to submit an audio task and access its results. Provide accurate account information, protect your login credentials, and do not use another person's account without permission. If you suspect unauthorized access, stop using the affected session and report it through the support channel once that channel is provided.
Task access is restricted by account ownership. Temporary playback and download links should be treated as private: someone who receives a valid link may be able to use it until it expires.
You must have the rights and permissions needed to upload, process, and use the recordings you submit, including any applicable copyright, performer, voice, and privacy permissions. Do not submit content that unlawfully infringes another person's rights or violates applicable law.
Uploading a recording does not transfer your ownership of it to Audio Studio. You give the service permission to store, process, and transmit that recording and its outputs only as needed to provide the requested feature, including through the stated storage and processing providers. This draft does not grant a separate, perpetual licence to use your recordings for model training.
Audio separation does not create ownership or a licence over someone else's music, recording, or voice. A generated instrumental or isolated vocal may still be subject to third-party rights. You are responsible for obtaining any permissions needed for your intended publication, distribution, or commercial use.
The current audio upload accepts MP3 and WAV files up to 50 MiB. The browser instrumental mix currently supports tracks up to five minutes. This mixing limit is not a server-enforced duration limit for every separation task. Formats, limits, and local trial quotas may change during development; check the workspace before submitting.
Video processing accepts H.264 video in MP4 or MOV files and VP8 or VP9 video in WebM files, with a supported audio track, up to 50 MiB and 300 seconds. Audio is extracted as a 128 kbps MP3 for separation. The browser retains the original picture and encodes the separated voices into the output video. First use loads a local processing component of about 8 MB; extraction and export depend on your device's available memory, processing power, and browser support. A supported file extension alone does not guarantee that a file can be processed or played.
Local tasks have a one-hour processing window. After that window, the service will stop advancing an unfinished task once any active save or query finishes. This timeout does not cancel a job already submitted to the processing provider or guarantee that its charges stop.
Separation quality depends on the recording and model. Results may contain residual music, missing sounds, distortion, or other artifacts. Video music removal aims to retain human voices, including speech and singing; environmental sounds and sound effects may be removed. Perfect isolation, unchanged speech, and suitability for a particular use are not guaranteed. Keep your own copy of the original and review outputs before using them.
Processing depends on third-party availability and may fail or be interrupted. The current workflow relies on the workspace checking progress to finish saving results; keep it open or return to the task to resume checking. Closing the page, removing the selected file, or starting a different preview does not cancel a submitted provider task and may not stop provider costs incurred by the operator.
Completed audio results can be accessed through the service for 24 hours. Individual signed links last up to 15 minutes, may expire sooner near the end of that access period, and can be refreshed by the task owner while access remains available. Download any results you want to keep before access expires.
Video exports are created in browser memory and are not stored by the service. A full page reload cannot recover the original video or its local export. The same account can resume checking a saved audio task while it remains accessible, but this does not restore the video. Download local video exports before refreshing or leaving the page.
These access periods do not promise physical file deletion. The media cleanup routine defaults to a non-deleting preview and requires separate activation and scheduling for actual deletion. Its scope, waiting periods, retries, and remaining retention details are described in the Privacy Policy. It does not delete provider-held copies or your downloads. Self-service account and audio deletion is not yet available. Do not rely on the service as a permanent archive.
The current local audio separation trial does not charge users or deduct credits for a separation task. The operator may incur third-party processing and storage costs when real processing is enabled.
Public pricing, subscriptions, credit rules, cancellation, failed-task charges, and refund policies are to be provided. If paid access is introduced, the applicable price and conditions must be shown before purchase. These draft terms do not impose an automatic renewal or a blanket no-refund rule. Mandatory consumer rights remain applicable.
Do not attempt unauthorized access to accounts or files, bypass security controls or usage limits, upload malicious files, or disrupt the service. Use the service in accordance with applicable law and the permissions described above.
Access may need to be restricted for a credible security concern, unlawful use, or a material violation of these restrictions. Any public suspension, termination, and appeal procedures must be finalized before launch, with notice where appropriate and subject to applicable law.
This development service may change or be unavailable during testing. Any launch warranties, liability provisions, and dispute procedures are to be provided after the operator and markets are confirmed. This draft does not exclude liability that cannot legally be excluded, remove mandatory consumer protections, or require binding arbitration.
These drafts will be revised before public launch. Final terms must identify their effective date and explain the notice or acceptance process for material changes. Support contact details are to be provided.
See the Privacy Policy for the current account, file, and third-party data workflow.