04 - AI Content Terms

Last updated: 2026-09-23

These AI Content Terms form part of the Terms of Service and govern ownership, rights and responsibilities for everything you put into PyxelJam.Studio ("PXLJ") and everything it produces.

We have written this to be useful rather than defensive. Some of what follows is a promise. Some of it is an honest statement that nobody in this industry can promise you certainty yet, and we would rather you know that now than find out in a distribution deal.

---

1. Definitions

• Inputs - anything you supply: scripts, synopses, prompts, uploaded images, reference photographs, character sheets, location photos, audio, on-camera recordings, LoRA training material, notes and settings.
• Outputs - anything the Service produces from your Inputs: storyboard images, character sheets, video clips, generated audio and foley, transcripts, and assembled exports.
• Model Providers - the third-party AI companies whose models actually perform the generation. PXLJ does not train, host or run its own models.

2. You own your Inputs

2.1 Your Inputs are yours before you upload them and remain yours afterwards. We claim no ownership of them.

2.2 You grant us a limited, non-exclusive, worldwide, royalty-free licence to use your Inputs solely to operate the Service for you. That means: storing them, displaying them back to you and to people you deliberately share with, transmitting them to Model Providers so a generation you asked for can run, and creating derived working files (thumbnails, poster frames, normalised audio, re-encoded video, character sheets, extracted stills).

2.3 That licence exists only for as long as we need it to provide the Service, and it ends when you delete the content or your account, subject to the backup and legal-retention periods in the Privacy Policy.

2.4 The licence is not a licence to publish, market, resell, showcase or train on your Inputs. See Section 8 and Section 9.

3. You own your Outputs

3.1 As between you and PXLJ, you own the Outputs generated in your account. To the fullest extent we are able to grant it, we assign to you all right, title and interest we may have in them.

3.2 We claim no ownership stake, no royalty, no revenue share and no continuing licence to reuse your Outputs for our own purposes beyond operating the Service for you.

3.3 Commercial use is permitted on every tier. You may use Outputs in commercial productions, client work, advertising, films, series, microdramas and social content, and you may monetise them, subject to the rest of this document and to the Acceptable Use Policy.

3.4 There are two honest caveats to Section 3.1, and they are Sections 4 and 5.

4. Caveat one: copyright in AI output is legally unsettled

4.1 "You own it as between you and us" is a real promise, but it is a promise about us. It is not a promise that the law will grant you an enforceable copyright in a given Output. That question is genuinely unresolved and it differs by country.

4.2 The United States. The U.S. Copyright Office has taken the position that copyright protects only works of human authorship, and that material generated purely by an AI system in response to a prompt is not itself copyrightable, however elaborate the prompt. Its 2023 registration guidance and its 2025 report on copyrightability say that a work containing AI-generated material can still be registered, but only for the parts a human authored: the selection, coordination and arrangement of the material, and any human-authored content within it. This position was reinforced in Thaler v. Perlmutter.

4.3 The European Union. EU copyright protects a work that is the author's own intellectual creation reflecting free and creative choices. There is no EU-level decision squarely on prompt-generated imagery yet. The prevailing academic and national-office view is broadly similar to the U.S. one: human creative contribution is what attracts protection.

4.4 The United Kingdom. UK law has an unusual provision for "computer-generated works" with no human author, giving a 50-year term to the person who made the arrangements necessary for creation. Its application to modern generative AI is untested and has been under active review.

4.5 What this means in practice for you.

• You can still use and sell the work. Uncertain copyright is not the same as no rights. Nothing here stops you distributing, licensing or monetising an Output.
• Enforcement against copycats may be weaker. If someone lifts a purely machine-generated frame from your film, your infringement claim may be harder in some jurisdictions than it would be for live-action footage.
• Your human contribution counts, and PXLJ is built around it. Your script, your shot list, your scene structure, your directing choices, your character design decisions, your take selection, your edit and your assembly are all human authorship. A finished PXLJ production is a compilation of many human creative choices, and that is generally protectable even where an individual frame might not be.
• Keep your working record. Your project history in PXLJ - drafts, shot lists, prompt iterations, rejected takes, the takes you chose - is evidence of human authorship. If you plan to register a work or license it into a production where chain-of-title is examined, keep that record and be ready to disclose which parts are AI-generated. The U.S. Copyright Office requires that disclosure on registration.
• Distributors, broadcasters and insurers are asking. If your work is going into a chain-of-title review, an errors-and-omissions insurance application or a platform with AI disclosure rules, expect to be asked what was generated and how. Answer accurately.

4.6 This is not legal advice. If a specific production depends on a specific rights position, take advice from a lawyer in the relevant jurisdiction.

5. Caveat two: Model Provider terms pass through

5.1 PXLJ does not run its own models. When you generate, your Inputs are sent to a Model Provider, that provider's model produces the Output, and that provider's terms of service apply to that processing and to the Output it produced, in addition to these terms.

5.2 The Model Providers currently used by PXLJ are:

FAL.ai routes to models from ByteDance (Seedream, Seedance, DreamActor, OmniHuman), Black Forest Labs (FLUX), Lightricks (LTX), Kuaishou (Kling), MiniMax, Alibaba (Wan) and Google (Gemini Omni). Which model handles a given job depends on the model you select and, where a fallback applies, on availability.

5.3 Today, the providers we use grant users broad rights to model outputs, including commercial use. That is why PXLJ can make the ownership statement in Section 3.

5.4 But those terms are theirs, not ours, and they change. A provider can change its output licence, add a restriction, deprecate a model or withdraw a capability, and we cannot prevent it. We do not control those terms and we do not warrant them.

5.5 We will pass on what we can. Where we become aware of a Model Provider change that materially affects your rights in Outputs already produced through PXLJ, we will tell affected customers by email or in-product. We cannot guarantee we will learn of every change.

5.6 If a specific production depends on a specific model's licence terms, check that provider's current terms before you commit, or ask us at support@pyxeljam.studio and we will point you to them.

6. No warranty of uniqueness or non-infringement

6.1 Outputs are not unique. Generative models produce statistically similar results for similar prompts. Another user, on PXLJ or on any other platform using the same models, may generate an Output that is very similar or occasionally near-identical to yours. We do not, and cannot, guarantee uniqueness or exclusivity of any Output.

6.2 We do not warrant that Outputs are free of third-party rights. A model may produce something that resembles an existing work, a trademark, a distinctive style, a recognisable character, a real logo or a real person's face, whether or not you asked for it. We do not review Outputs for infringement and we cannot.

6.3 No indemnity. Unlike some large model vendors, PXLJ does not offer a copyright indemnity for Outputs. You use them at your own risk.

6.4 Your responsibility before you publish. You are responsible for clearing Outputs before commercial use: checking for recognisable trademarks, real logos, protected characters and unintended likenesses, and for any clearance your distributor, broadcaster or insurer requires.

6.5 Accuracy. Model output can be wrong. Generated text, transcripts, shot suggestions and continuity notes may contain errors and invented detail. Do not rely on any Output as a statement of fact without checking it.

7. Likeness, faces and voices - your responsibility

This section matters more than any other in this document, and it applies to every reference photograph you upload and to every recording you make with the Acting feature.

7.1 What you are responsible for. Before you upload or record any material depicting a real, identifiable person - their face, their body, their voice - you must hold all rights and consents needed to use it for the purpose you are using it for, including generating derivative synthetic media from it. That includes:

• consent from the person, informed and specific enough to cover AI generation;
• any release, performer agreement or union clearance that applies;
• a lawful basis under data protection law, because a face or voice recording is personal data and in many jurisdictions biometric or otherwise sensitive data;
• rights of publicity and likeness, which in the United States are a matter of state law and vary sharply: California protects both a statutory and a common-law right of publicity (Cal. Civ. Code sections 3344 and 3344.1, the latter covering deceased personalities), New York protects name, portrait, picture and voice (N.Y. Civ. Rights Law sections 50 and 51, extended in 2020 to digital replicas of deceased performers), and Tennessee's ELVIS Act expressly reaches AI voice and likeness cloning. A growing number of states have added synthetic-media and digital-replica statutes, and the federal TAKE IT DOWN Act reaches non-consensual intimate imagery including AI-generated depictions;
• biometric privacy statutes, notably Illinois' BIPA and Texas' CUBI, which regulate face and voice data specifically and, in Illinois, carry a private right of action;
• union and guild terms where a performer is covered - the SAG-AFTRA agreements now contain specific digital-replica consent requirements;
• personality and image rights outside the US, which differ by country and in several European jurisdictions are protected independently of copyright.

7.2 This applies to you as well as to other people. If the face or voice is your own, you consent by uploading it. If it is anyone else's - a friend, an actor you hired, a client, a member of the public in the background - you need their permission.

7.3 The Acting feature specifically. When you record or upload a voice take or an on-camera take and use it to drive a shot, that recording is transmitted to a third-party model provider and used to animate a character. The resulting clip carries that performance. Before you record someone else, tell them plainly what will happen to the recording and get their agreement.

7.4 Voice cloning specifically. Creating a synthetic voice from a person's recorded speech requires that person's explicit consent. Cloning a voice without consent is prohibited by the Acceptable Use Policy and is unlawful in a growing number of jurisdictions.

7.5 Character library entries. A character built from a real person's photographs inherits every obligation above, and the entry persists in your account across projects. If consent is withdrawn, delete the character and the material derived from it.

7.6 We do not verify consent. We have no way to know whether you have it. We rely on you, and you indemnify us if you did not (see the Terms of Service, Section 14).

8. Training - our position and theirs

8.1 PXLJ does not train AI models on your content. We do not use your Inputs or your Outputs to train, fine-tune, evaluate or improve any model, our own or anyone else's. We do not sell or licence your content as a training dataset. We do not use your content to build features for other customers.

8.2 Human review is limited. We do not routinely read your scripts or watch your generations. Our staff may access your content only where you ask us to for support, where we are investigating a specific and credible Acceptable Use Policy breach or security incident, or where the law requires it.

8.3 Upstream providers have their own policies, and we do not control them. Your content passes to a Model Provider when a generation runs, and what that provider does with it is governed by their terms.

• Providers on business or API terms generally commit not to train on API inputs and outputs by default. That is the norm across the providers we use, and it is one of the reasons we use their API tiers rather than consumer products.
• Providers typically retain inputs and outputs for a short period for abuse monitoring and safety review, even where they do not train on them.
• Policies differ per provider and per model, and they can change.

8.4 What this means for you. If you are working on material that must never leave your control - an unreleased script, a client's confidential campaign, a person's identifiable face under a strict data agreement - understand that running a generation transmits it to a third-party provider in the United States. If that is not acceptable for a specific project, do not run that material through the Service.

8.5 LoRAs and custom models. Where you train a custom model (LoRA) on your own material through the Service, the training material and the resulting model are yours, used only for your account, and not shared with other customers.

9. Confidentiality of your projects

9.1 Your projects are private to your account by default. Other customers cannot see them.

9.2 You control sharing. Review shares and public shares are token-gated links you create deliberately, and you can revoke them at any time. Anyone holding a live link can see what it points at, so treat a share link as the credential it is.

9.3 We will not publish, showcase, screenshot or use your work in our marketing without your prior written permission, asked for specifically.

10. Content filtering by providers

10.1 Model Providers run their own safety filters. A generation may be refused by a provider even where it complies with our Acceptable Use Policy - a violent scene, an injury, a body in swimwear, a real person's name - because the provider's classifier read the prompt differently.

10.2 We surface those refusals and offer a reword-and-retry path, and the Service phrases machine-authored prompts to reduce false refusals. We cannot override a provider's decision.

10.3 Where a job reached the provider before being refused, the underlying cost may already have been incurred. See the Terms of Service, Section 6.7, for when credits are and are not refundable.

11. Attribution and disclosure

11.1 Attribution and watermarking. We never require you to credit PXLJ.

• On Standard and Pro, your Outputs carry no PXLJ mark of any kind.
• On Starter, which is a one-time trial rather than a recurring free plan, an assembled export carries the PyxelJam wordmark in the bottom-right corner. Individual takes and still images you download are not marked; the mark applies to assembled exports only. Upgrading to a paid tier and re-exporting produces a clean file.

11.2 You are responsible for complying with any AI disclosure obligation that applies to your work. These are growing quickly: the EU AI Act's transparency obligations for deepfakes and synthetic media (Article 50), platform disclosure rules on YouTube, TikTok, Meta and elsewhere, election-advertising rules in many countries, and broadcaster and distributor requirements.

11.3 The Acceptable Use Policy separately prohibits presenting synthetic media as genuine footage of real events.

12. Changes

We may update these AI Content Terms as the law and the model landscape develop. Material changes are notified as described in the Terms of Service, Section 15.

---

Questions: support@pyxeljam.studio