How We Use AI

1. Harness, not provider

SceneCrew is not an AI model provider. We operate no AI model. The platform is a production harness that turns a creative brief into a finished video by routing content through AI models run by other companies, called AI Providers. Our software does the following categories of work:

• Collects your creative instructions (the brief, the Director choice, format, constraints);
• Builds the structured prompts the Director system sends at each stage — story, cast, script, scenes, final cut;
• Sends those prompts to the AI Provider you have selected, either with credentials we hold (Managed mode) or credentials you have provided (BYOK mode);
• Receives the AI Provider's response, stores it for you, reviews it through a second Provider call where our Director protocol calls for a gate review, and lets you iterate or approve;
• Assembles the approved stages into a final video using ffmpeg-based composition — this step is deterministic and contains no AI.

2. AI Providers we route to

The list below is accurate as of the “Last updated” date. We add and retire Providers as the field moves. Changes are reflected here and on the Privacy Policy subprocessor list.

Fourth-party model vendors — for example, Kuaishou (Kling) or Alibaba (Wan) — host their models through fal.ai. Their terms flow through fal.ai's published policies. Please read your chosen Provider's published terms before relying on their Outputs.

3. How your content flows

A typical project runs like this:

1. You submit a brief. The brief is stored in our database on our hosted infrastructure (Germany, Contabo).
2. When you trigger a stage, our server composes a prompt using your brief, the Director identity you chose, and any stage context (prior Outputs the Director has approved). The prompt is sent to the AI Provider you have selected.
3. The Provider processes the prompt on its own infrastructure, typically in the United States or an EU region depending on the Provider, and returns a response. Response handling follows the Provider's published terms.
4. We receive the Output, save it to your project workspace on our servers, and (for gate stages) run a second Provider call that reviews the Output against your brief and the Director's Constitution.
5. You review, edit, approve, or regenerate. Approved stages are composed locally with ffmpeg into the final video.

In Managedmode, the AI Provider contract is between us and the Provider; we are the Provider's customer; you are ours. In BYOKmode, your API keys mean the Provider contract is between you and the Provider directly; we pass your Inputs through under your key and return the Provider's response. See the Terms of Service, section 6.

4. Your content and AI training

Our position: we do not operate any AI model. We do not have any AI model. We do not train on Your Content. Full stop.

Third-party Providers:whether a Provider uses data you send through SceneCrew to train or improve its own models is governed by that Provider's terms and changes over time. Our operational posture is:

• Where a Provider publishes a no-training, zero-retention, or enterprise-equivalent option, we prefer it for Managed mode and configure our integration to use it by default where we can;
• Where a Provider's consumer tier opts data into training by default and no opt-out exists, we avoid using that tier for Managed traffic;
• In BYOK mode, your choice of tier at the Provider controls — you should review your account settings with each Provider you use.

The concrete configuration is operational detail that can change — we will not guarantee it in the Terms. What we guarantee is that we will publish a truthful summary here, update it when Provider policy changes, and not silently route your data to training paths we have not disclosed.

5. What AI Outputs are — and are not

AI Outputs are probabilistic. They are built by statistical models trained on large collections of text, images, video, and audio. You should expect, and verify for, the following:

• Inaccuracy.Outputs can state false facts (“hallucinate”), misquote, or misattribute.
• Bias. Outputs can reproduce stereotypes or cultural biases embedded in training data.
• Similarity to protected works. Outputs can closely resemble copyrighted works, trademarks, or the likenesses of real people. You are responsible for checking before use.
• Refusals and filters. Providers apply safety filters. A prompt you believe is benign may be refused; an Output you have already received may be redacted or modified.
• Quality drift.Provider models change. Today's Output may not match yesterday's even from the same prompt with the same settings.
• No professional advice. Outputs are not legal, medical, financial, safety-critical, or any other regulated advice. Do not treat them as such.

You must review every Output before relying on it, sharing it, or publishing it. See the Terms section 10 and the Content Rights & Consent Policy.

6. AI transparency obligations

6.1 Labelling AI-generated content (EU AI Act)

Under Article 50 of Regulation (EU) 2024/1689 (the EU AI Act), anyone who deploys an AI system that generates or manipulates image, audio, or video content constituting a deepfake must disclose that the content has been artificially generated or manipulated. Anyone who generates or manipulates text on matters of public interest must similarly disclose it unless the content has been reviewed by a human editor exercising editorial responsibility.

You are the deployer of the Output you produce using SceneCrew. It is your responsibility to comply with Article 50 and equivalent laws. Our position is that all video output from SceneCrew should be clearly labelled as AI-generated when you publish, share, or commercialise it. Where we are able to embed provenance signals (for example C2PA content credentials) or a visible watermark, we will — and we reserve the right to do so without notice in order to meet our own compliance obligations.

6.2 No identifying real people without consent

Whether a transparency label is present or not, using the Service to generate deceptive content about real people is prohibited. See the Content Rights & Consent Policy.

7. Our internal use of AI

In addition to the generative AI that powers the Service, we use AI-adjacent tooling to help run the business. Where this processes personal data, the legal bases are covered by the Privacy Policy.

• Engineering copilots for code review and troubleshooting. We do not feed Your Content into these tools.
• Prompt defaults and Director personas — system prompts we author and store in our database. These control the behaviour of the AI Providers we call. You may override them in your account via the prompt editor; those overrides are attributed to you in our database and never shared with other accounts.
• Automated content-safety checks applied by Providers (and any complementary checks we run) may block or flag generations that appear to violate the Acceptable Use Policy. Decisions are reviewable on request.

8. Your responsibilities when using AI

• Read and comply with the Provider's terms you have selected. This is especially important in BYOK mode.
• Do not upload content you do not have the rights to. See the Content Rights & Consent Policy.
• Review every Output before you publish it. Verify accuracy, check for likeness of real people, check for protected works, check for brand insertion you did not approve.
• Label AI-generated content where a law, platform policy, or audience expectation requires it.
• Respect the safety behaviour of Providers. Do not attempt to bypass safety filters; do not share methods of doing so with others on the Service.
• Do not use Outputs in regulated high-risk contexts (employment, credit, insurance, law enforcement, critical infrastructure, medical) without prior written agreement from us and an assessment that complies with the relevant law, including the EU AI Act.

9. Reporting AI safety issues

If an Output you receive appears to cause a safety problem — for example an unexpected deepfake of a real person, or content that appears to violate the Acceptable Use Policy — please report it to abuse@scenecrewai.com. We will investigate and, where appropriate, raise the matter with the AI Provider. If you believe the Output infringes your intellectual-property rights, use the IP Takedown process.

10. Changes to this page

This page reflects our current integrations and operational posture. The list of Providers and modes of access will change as we add or retire capabilities. We will update this page when that happens and reflect material changes in the Privacy Policy subprocessor table.