Responsible AI

Last updated: May 30, 2026

1. Our commitments

Lumino is operated by Lumino Labs LLC and integrates third-party AI models from Decart, fal.ai, HeyGen, ElevenLabs, and Google AI. We make the following commitments to our users, the people in our users' content, and the providers we build on:

  • Every generation is tied to an authenticated user account. There is no anonymous public generation endpoint.
  • We do not allow impersonation of real people without their consent, and our product surfaces are designed not to invite it.
  • We respect content-provenance markers applied by upstream models. Where we have approval to deliver unwatermarked output, we mark content as AI-generated by other means.
  • We act on credible abuse reports within a reasonable timeframe, and we share abuse signals with upstream providers when there is a safety reason to do so.
  • We do not sell user uploads or generations to train third-party AI models outside what the requested generation requires.

2. How content is generated

Generations on Lumino are produced by third-party AI models. We do not train our own foundation models. Each tool calls a specific upstream model:

  • Avatar Video (Standard): Kling Avatar v2 via fal.ai.
  • Avatar Video (Premium): HeyGen Avatar IV via fal.ai.
  • Voice Clone & Custom TTS: ElevenLabs Instant Voice Clone and Multilingual TTS.
  • Photo Studio: Flux Kontext Pro and FLUX.2 Pro Edit via fal.ai.
  • Live Transform: Decart real-time video models.
  • AI Writer: Google Gemini 2.0 Flash.
  • Document Generator: server-side jsPDF (no AI text generation on the output document itself; user-provided fields).

Each upstream provider has its own acceptable-use policy. We respect those policies in addition to our own.

3. Content provenance & labeling

We treat AI-generated content as distinct from human-recorded content and label it accordingly.

  • Where an upstream model applies a visible watermark or provenance signal, that signal remains in place by default and may not be removed by users.
  • For Live Transform specifically, where we have an approved removed-watermark agreement with Decart, we mark output as AI-generated through other means (in-product notice, session metadata, and post-session disclosure).
  • We support C2PA-style provenance metadata where the upstream pipeline emits it.
  • Voice clones and avatar videos generated on Lumino remain identifiable as AI-generated through the upstream provider's own provenance systems.

4. Consent

We rely on attestation at the point of generation, paired with policy enforcement when attestation turns out to be false.

  • Voice cloning: by submitting an audio sample, you attest that the speaker is you or has given documented permission. Cloning a third-party voice without consent is grounds for immediate termination.
  • Live Transform: each session requires an explicit on-screen consent confirming the subject is you or has consented to be transformed and broadcast.
  • Avatar Video and Photo Studio: reference images you upload must be of yourself or a person who has consented. Lumino does not allow reference images of public figures without consent.
  • Tax, KYC, banking, broker, and document tools: simulated only, and may not be used to misrepresent real identities or transactions.

5. Moderation

Moderation is layered. We rely on upstream model safety filters, our own policy enforcement, and human review of reports.

  • Upstream models apply their own content filters (NSFW, violence, prohibited categories). Lumino does not attempt to bypass these.
  • Prompts and uploads are subject to file-type validation, magic-byte checks, and basic content checks before being passed to the upstream model.
  • Every generation is associated with an account ID, timestamp, and (where applicable) prompt or reference for review.
  • Reported content is reviewed by a human. Confirmed violations result in content removal, account action, credit forfeiture, and reporting where appropriate.

6. What we will not build

Some capabilities are obviously high-risk and we choose not to offer them, regardless of demand:

  • Live Transform presets, avatar templates, or voice presets that target named real people.
  • Tools designed to bypass another platform's security, authentication, or terms of service.
  • Tools that generate sexual content involving any person who has not given explicit consent, and unconditionally no content involving minors.
  • Tools that produce real (as opposed to clearly-simulated) financial, travel, identity, or legal documents.
  • APIs that allow third parties to generate content against our account without per-end-user accountability.

7. Enforcement framework

When abuse is identified, our response scales with severity:

  • Minor violations: content removal and a warning. The user's account remains active.
  • Repeat or serious violations: temporary suspension, credit forfeiture, deletion of relevant generations or voice clones.
  • Severe violations: permanent termination, deletion of all account content, and reporting to upstream providers, affected platforms, and law enforcement where appropriate. Examples include CSAM, non-consensual intimate imagery, credible threats of violence, and use of our tools to commit fraud against identifiable victims.

8. Data & training

We do not use your uploads, prompts, voice samples, or generated content to train third-party AI models outside what the upstream provider requires to perform the requested generation. Generated videos, photos, and TTS audio are retained for up to 7 days on object storage and then permanently deleted. Voice clones are retained until you delete them. KYC images in the simulated banking and brokerage portals are retained for the lifetime of the simulated account and deleted with it.

9. Reporting and takedowns

See our Trust & Safety policy for the full prohibited-uses list and the reporting process. For abuse, write to abuse@lumino.work. We aim to respond within 3 business days.

10. Changes

We update this page as the product and the wider regulatory environment evolve. The "Last updated" date at the top reflects the most recent revision.

11. Contact

For Responsible AI questions, email support@lumino.work.

Operated by Lumino Labs LLC, registered in Delaware, United States.