AI Features
Terms

These AI Features Terms supplement the VLStudio Terms of Service and apply whenever you use an AI feature in VLStudio Desktop: AI chat, AI Visuals, cloud captions, the YouTube Coach, the generate-video pipeline, and the MusicGen music generator. All paid AI features are gated behind the VLS-PRO subscription and access is withdrawn if the subscription lapses.

1. What Is Sent, and to Whom

VLStudio Desktop's AI features are routed through the VLStudio Render backend, which forwards requests to Google Gemini and to Groq. This section states exactly what data leaves your device for each feature, so there is no ambiguity about scope.

• AI chat: your prompt, a serialized snapshot of your project timeline (clips, tracks, markers, caption text, and transforms), and your chat history are sent to the VLStudio Render backend, which forwards them to Google Gemini.
• AI Visuals: raw JPEG frames extracted directly from your footage, up to six frames per turn, are sent to Google Gemini so the assistant can see what is in the shot. These are actual images from your footage, not descriptions of it.
• Captions: three paths, different handling. One runs fully on-device using a local Whisper model (the model file itself may download from Hugging Face the first time you use it, a one-time model transfer, not your footage or audio). One sends audio to Google Gemini via our backend. One sends audio directly to Groq. The app indicates which path is active.
• YouTube Coach: your video transcript and channel context are sent to Google Gemini using your own Google Gemini API key, which you supply. This exchange is directly between you and Google.
• Server-side interaction logging: when enabled on the backend, each AI turn (your message, the full timeline context, and the raw model response) is persisted server-side as a dataset. See Section 5 for status.

2. Frames May Contain Identifiable People, Places & Third-Party Footage

The raw JPEG frames sent for AI Visuals are pulled directly from your footage. If your footage shows people, it may show identifiable faces. If it shows a location, that location may be identifiable. If your footage includes third-party material, that material is included in the frames sent to Google Gemini exactly as it appears in your project.

3. Who Owns AI Output

You own the AI-generated edits, captions, and analysis produced when you use VLStudio's AI features on your own content, subject to the underlying rights in whatever footage, audio, or material you fed into the AI. AI output is only as clean, rights-wise, as the inputs you provide.

You warrant that you hold, or have obtained, the necessary rights to any footage, audio, images, or other material you submit to an AI feature, and that submitting it to Google Gemini or Groq for processing does not infringe any third party's rights. If you do not hold those rights, do not submit the material to an AI feature.

4. Google's and Groq's Own Terms Govern What You Send Them

When you use AI chat, AI Visuals, or the two cloud caption paths, your content is submitted to Google Gemini or Groq, and their own terms of service and privacy policies govern their handling of that content in addition to this document. We do not control, and cannot override, how Google or Groq process content once it reaches them.

The YouTube Coach feature is different: it uses your own Google Gemini API key, so your transcript and channel context go directly from your device to Google under Google's consumer terms, the same as any other use of your own API key. VLStudio does not receive, see, or store that data at any point.

5. AI Provider Retention

Whether Google Gemini and Groq's contracted tiers permit VLStudio to make a no-training or no-retention representation about content sent through those paths, and whether server-side interaction logging currently ships enabled, is: [[AI_PROVIDER_RETENTION]].

We do not represent, one way or the other, that your prompts, timeline data, frames, or audio are excluded from provider-side training or retained or discarded on any particular schedule. This section will be updated once that status is confirmed. Until then, treat content submitted to AI features as potentially retained by the receiving provider under that provider's own terms.

6. AI Long-Term Memory

The AI assistant has a long-term memory feature. It extracts and remembers your name, role, goals, current project, and stated preferences across sessions, so it can personalize its responses to you without you having to repeat context every time.

You can ask the assistant to forget what it has stored, or contact us to request erasure, see Section 11.

7. Acceptable Use of AI Inputs

You may not submit unlawful or infringing content to any AI feature. This includes, without limitation, content you do not have the rights to, content that infringes a third party's intellectual property, content that violates a third party's privacy or publicity rights, and content that is illegal to possess or distribute in your jurisdiction.

This restriction also applies to two additional features:

• The generate-video pipeline: this feature ingests content from Telegram and results from web searches as part of generating video output. Inputs and outputs of this pipeline are subject to the same acceptable-use restriction: do not direct it at, or use it to reproduce, unlawful or infringing material.
• The MusicGen generator: generated music output is subject to the same restriction, and separately, MusicGen's underlying model weights carry a non-commercial license (CC-BY-NC-4.0), which limits how output from this specific generator may be used regardless of the acceptable-use rule.

Whether the generate-video pipeline and MusicGen are end-user-facing features or internal-only tooling is: [[TELEGRAM_MUSICGEN_STATUS]]. Until this is confirmed, we treat both as live, user-facing, experimental features, and they are additionally subject to our Beta / Experimental-Features Disclaimer.

8. EU AI Act Transparency

The EU AI Act distinguishes between an AI "provider" and an AI "deployer," with different obligations attached to each role. VLStudio's server-side interaction-logging feature, described in Section 1 and Section 5, may place VLStudio in the role of an AI provider for the logged interactions, rather than only a deployer of Google's and Groq's models. This classification is being assessed. We will update this section once that assessment is complete.

9. No Warranty on AI Accuracy

AI output is informational and assistive. It is not guaranteed to be accurate, complete, or fit for any particular purpose, and it depends on the uptime and behavior of third-party providers we do not control. See our No-Reliance Disclaimer for the full scope of this limitation, including its application to the YouTube Coach's recommendations.

10. Related Documents

• Biometric Data Notice: facial geometry and voiceprint exposure from AI Visuals and the caption paths.
• No-Reliance Disclaimer: the informational-only status of AI output and platform guidance.
• Beta / Experimental-Features Disclaimer: the generate-video pipeline, MusicGen, and other unfinished surfaces.
• Privacy Policy: the full data-flow disclosure for AI features as personal-data processing.

11. Contact

Questions about these AI Features Terms: [[CONTACT_EMAIL_LEGAL]] (interim: vlstudiopartners@hotmail.com).