Hydras
Rights Matrix
Last verified 17 September 2026
Training
Do the providers use your content to train their models?
The question people actually mean is: could my unreleased production end up inside someone's model? Below is what each provider's terms permit, not what we hope they do.
| Provider | What it does in Hydras | Trains on your content | On what basis | Source |
|---|---|---|---|---|
| OpenAIGPT-6.1 Sol, GPT Image 2 | Script and episode preparation, shotlists, storyboard frames | No | API content is "not used to train or improve OpenAI models (unless you explicitly opt in)". We have not opted in. | Read ↗ |
| GoogleGemini, Vertex AI | Shot and video prompts, character continuity, still images | No | Contractual: "Google will not use Customer Data to train or fine-tune any AI/ML models without Customer's prior permission or instruction." | Read ↗ |
| AnthropicClaude | In-product assistant | No | Contractual: "Anthropic may not train models on Customer Content from Services." | Read ↗ |
| Kling AIKling 3 | Video generation | No | Contractual: "We will not use … your data to train, retrain, or otherwise improve the Services or any other product." | Read ↗ |
| ElevenLabs | Voiceover and character dialogue | No | Training is switched off on our account, and their terms hold them to that setting for as long as it stands. | Read ↗ |
| Topaz Labs | Delivery upscaling | No | Privacy Policy §3B, which EULA §2 points to: "Topaz Labs will not use Your media content (including images and videos) to train its artificial intelligence models without Your prior consent". We have not consented. The same clause lets them use de-identified, aggregated or synthetic data derived from media to improve their algorithms, provided it cannot be linked back to you. Their public statement goes further: "We never train models using customer data or content." | Read ↗ |
| BytePlusSeedance — not enabled | Alternate video model, switched off | No | Training requires a separate opt-in agreement. We have never accepted it, and the model is not enabled for customer projects. | Read ↗ |
| fal.ai | Optional storyboard camera-angle re-render | No | "Company will not use Client Content to create, train, develop (directly or indirectly) Company's products or services." Their agreement also treats what you send as confidential information, and restricts them to our written instructions. | Read ↗ |
The short version. Every provider that generates your material is committed not to train on it. Most say so in a single clause; one reaches the same place through a purpose-limited licence and a data processing agreement, which the table explains rather than glosses over. Where a commitment rests on an account setting rather than a contract clause, the table says that too.
We choose providers on this basis. A model whose terms claim the right to train on what you send is not one we will use, however good its output. That has already cost us one storyboard model, and we would make the same call again.
Ownership
What rights do the providers keep in what you generate?
The second question is quieter but matters more at delivery: if a broadcaster asks who owns this footage, is there anyone else in the chain?
| Provider | Who owns the output | Royalty claimed | What they do keep | Source |
|---|---|---|---|---|
| OpenAI | Yours | None | Their models. They go further than most: "we hereby assign to you all our right, title, and interest, if any, in and to Output." | Read ↗ |
| Yours | None | Their models. "Generated Output is owned by the user/customer, and Google does not assert any ownership rights." | Read ↗ | |
| Anthropic | Yours | None | Their models, with an express assignment of their interest in output to the customer. | Read ↗ |
| Kling AI | Yours | None | Their models. Commercial use of the output is expressly "not restricted". | Read ↗ |
| ElevenLabs | Yours | None | Their voice models. "You retain all rights in and to your Output." Commercial use requires a paid plan, which we hold. | Read ↗ |
| Topaz Labs | Yours | None | Their software, algorithms and models. EULA §18.2.5 (§7 says the same): "You retain all ownership rights in Your Media, and Topaz does not and will not: (i) make any claim of copyright ownership related to Your original Media; and (ii) assert any right to royalties in relation to Your Media." | Read ↗ |
| BytePlus | Yours | None | Their models. Output is treated as customer data. | Read ↗ |
| fal.ai | Yours | None | Their models. "Customer owns and retains all right, title, and interest in and to the Customer Input", and no claim is made over the output. | Read ↗ |
The chain of title is clean. As at the date above, not one provider we use retains ownership of, or claims a royalty in, what you generate. Three go further and formally assign their interest to you. Nobody stands between you and your footage. Our own position — that we claim nothing in your output — is a commitment in the Terms of Service. What the providers claim is reported here from their documents, not promised by us.
About these tables
For every AI provider that receives Customer Content, this page quotes the clause in that provider's terms that governs training on your content and ownership of the output, with a link to the source and the date we last checked it. It is the evidence behind the commitments Warner Bros. International Television Production Finland Oy gives in Terms of Service Sections 5 and 7, which are the binding text. Where this page and those Sections differ, the Terms prevail.
Providers can change their documents. We re-check them regularly, and when a change affects either commitment, Section 5 of the Terms sets out what we do.
Limits
What these tables do not settle
Copyright protection. Whether an output is protected by copyright depends on national law and on the human creative contribution to it. No provider above claims your output, and Hydras keeps the record of the human direction behind each asset (Terms of Service Section 7).
"Not training on your content" is not "no data at all". Providers generally reserve the right to use anonymised, aggregated statistics about how their service is used — including figures derived from traffic — to develop their products. That is standard across the industry and is a different thing from your scripts, prompts and frames entering a training set, which is what the table above is about.
Not training is not the same as not keeping. Providers hold material for their own limited periods, mostly to detect abuse, and several reserve the right to review input and output for safety — which can mean a person looks at it. Section 5 of the Privacy Notice covers what reaches them and what does not, and our sub-processor list gives the period for each provider where we know it.
Where the material is processed is a separate question. Several of these providers operate outside the European Economic Area. Section 7 of the Privacy Notice deals with transfers.