Questions about the standard

Human Provenance in Film is a free, open standard for disclosing AI use in finished film and television productions. The producer selects one of three categories, No AI Used, Assistive AI or Generative AI, and records it in a signed declaration.

HPF exists so audiences can know how AI was used in the films and programmes they watch. It creates a common language for that information and a practical way for it to travel through the industry. When HPF is used, the declaration travels with the production through existing submission, acquisition and delivery documentation.

HPF does not tell organisations or audiences what decision to make. Each organisation applies its own policy to the information disclosed. The standard and its supporting documentation are published under Creative Commons Attribution 4.0, and it is free to make and use an HPF declaration.

The production company or lead producer records one classification for the finished production, and an authorised representative signs it. They can use the AI Disclosure Form or another document containing the required information.

The declaration should be based on the relevant information gathered from the people and suppliers responsible for creating, processing, supplying or integrating material that appears in the version being declared. Who produced a particular element does not change the classification: the highest applicable category across the finished production applies.

For step-by-step guidance on gathering the relevant facts, applying the taxonomy and completing the declaration, see Classify and declare your production. A lawyer or sales agent is not required.

Yes. A production can adopt HPF during production, in post-production, at acquisition or after release. For a production already completed, the producer, production company or another party authorised to make the declaration on the production's behalf may do so where it can establish the relevant facts. The declaration records the date it was made, not the production's release date.

A classification must not be inferred from a release date, detection output or incomplete catalogue information, and it must not be applied across a catalogue in bulk. Where the relevant facts cannot be established, the classification is unresolved. Missing or uncertain information is never No AI Used.

HPF classifies AI output reflected in the finished production an audience sees or hears. This includes output arising from development, pre-production, principal photography and post-production where it is embodied in the finished work.

AI use during development, planning, search or production management is outside the declaration where its output does not reach the released version. Influence alone is not incorporation. Human editing, repainting, rewriting or compositing does not by itself settle the classification. The question is what AI output, if any, remains or is embodied in the named released version. Classify and declare your production gives paired examples.

Two examples of unresolved cases are substantially rewritten generated dialogue and AI-generated or AI-segmented mattes, masks or similar control data that shape the delivered picture. Both are part of consultation question 4. Marketing and promotional materials are outside scope.

Ask the people and suppliers responsible for the released version what AI function was used, what output it produced, and whether that output remains in the version you are declaring. Ask about functions, not only product names: an AI feature in a programme does not by itself mean AI output appears in the finished work. Standard digital processes that do not use AI do not count. Classify and declare your production covers this in detail.

Two points remain open under consultation question 1: whether reasonable enquiry is the right basis for a No AI Used declaration, and whether some established machine-learning functions, such as certain denoising, tracking and upscaling tools, require Assistive AI disclosure. Reasonable enquiry is not a current HPF requirement.

A production receives one classification based on its highest category of AI involvement. If qualifying AI-originated content appears anywhere in the finished production, it is Generative AI regardless of the content's duration, prominence or proportion.

For example, a film with one AI-generated environment is Generative AI even if everything else was created by human crew. The declaration can record where and how AI was used so recipients have more context than the overall classification alone.

Apply the reconstruction test. The question is whether the AI reconstructed content already present in human-created or captured material, or fabricated content that was not there.

Reconstruction restores, cleans or repairs supplied material, for example de-flickering or artefact removal. This is Assistive AI. Fabrication invents material that was never captured or created, such as generating a background revealed by removing an object. This is Generative AI. Where a tool does both, the fabrication determines the classification.

What the tool did decides the treatment, not whether the feature is called restoration, enhancement or inpainting. Whether some established machine-learning functions of this kind require disclosure at all is part of consultation question 1, and difficult hybrid restoration cases are part of question 4. See the consultation questions.

Descriptors are provisional, optional terms, under consultation, that add factual detail about the type of AI output present, such as Altered performance, Digital replica or Generated music. They never change the classification. Altered performance may apply to Assistive AI or Generative AI; the others apply to Generative AI; none applies to No AI Used.

More than one descriptor may apply. AI-generated speech representing an identifiable actor, inserted into footage of their captured performance, may carry Digital replica, Generated voice, and Altered performance together. Generated writing spoken by an actor in the usual way carries Generated writing instead. Descriptors record what is present in the finished production, not whether consent or contractual permission was given. The taxonomy lists the current terms.

The level at which episodic productions should be classified remains under consultation. The current provisional position is that each episode can receive its own classification. Where a season or series is classified as a whole, it takes the highest category present in the episodes it covers. How the released-version rule applies to episodes is part of consultation question 4.

The declaration travels with the production in its submission, acquisition or delivery documentation. It can be supplied to a sales agent, distributor, broadcaster, platform, festival or other receiving organisation, and can be listed as a delivery item where the parties choose.

The signed declaration is sufficient; no technical integration is required. Organisations may also carry the classification in their own metadata or systems.

The producer who signs the declaration is responsible for its accuracy. HPF does not audit the production or review the evidence behind individual declarations.

A receiving organisation may ask for supporting information under its own policies or agreements.

One declaration covers one released version. A later released version, such as a director’s cut or a substantive localisation, has its own declaration, even where the facts are unchanged. Some localisation and episodic cases remain under consultation. The earlier declaration continues to cover its version, and HPF does not require notice to its recipients, although contracts or distribution arrangements may. Classify and declare your production lists the releases that normally do and do not need a new declaration.

A correction is different. If an error is found in a declaration, the producer issues a corrected declaration that names the one it replaces.

If the declaration has been incorporated into an agreement, the parties should also follow that agreement's process for making changes.

It means no HPF record exists. It does not establish whether AI was or was not used, and it is never equivalent to No AI Used. An organisation may describe the production as undisclosed under its own policy, but Undisclosed is not a fourth HPF classification.

An organisation can request an HPF declaration in its existing commissioning, submission, acquisition or delivery process and retain the signed form with the production's documentation. It can map the disclosed information to its own eligibility, rights, commissioning, acquisition or display policies without changing the meaning of the HPF classification. A festival or awards body, for example, can request it during submission or before selection and use the information within its own eligibility or programming rules.

An organisation asking about AI use across the whole production process may need additional information beyond the finished-work declaration. HPF does not require any particular policy or technical system and does not set eligibility or make programming decisions. Each organisation keeps its own policies, and if it uses HPF within its rules, those rules should explain what effect, if any, each classification has.

HPF is designed so the information recorded during production and delivery can ultimately be shown to audiences. The classification can be displayed in credits, programme notes, catalogue entries, content information pages or other audience-facing information. A recipient may also show applicable descriptors to provide more specific context about the AI-generated or AI-modified content present.

HPF does not prescribe one display format. The classification and descriptors should retain their defined meanings wherever they appear, and an absent declaration must not be displayed as No AI Used.

An HPF declaration is not a certification, approval or quality mark. It records what kind of AI output is present; it does not judge whether that use was appropriate. A declaration can be accurate in any of the three categories.

It also does not establish consent, rights or compliance with legal, contractual or collective-agreement requirements. Those matters remain governed by the relevant documentation and rules. An HPF declaration can sit alongside that documentation, but it does not replace it. An organisation may use HPF alongside its own compliance, assurance or certification processes; those processes remain separate.

No. HPF records AI output present in the finished production. It does not assess the material used to train an AI tool, or whether consent, licensing or compensation requirements were met in creating that tool. Model inputs and production outputs are separate questions; HPF classifies the latter.

No. HPF covers film and television, including studio, broadcaster, platform and independent productions. It is being introduced through existing independent-film workflows, but the classification does not depend on production scale, financing model or route to audiences.

The Mise En Scène Company holds the copyright and publishes the standard under Creative Commons Attribution 4.0. The published material is free for anyone to use and adapt under that licence.

MSC originated and currently maintains HPF. It intends to find a neutral long-term home for the standard, preferably within an existing standards body, but no arrangement or timetable has been decided. The amendment process and version history are published in GOVERNANCE.md.

Yes. HPF uses AI to support its work, including occasional analysis of consultation feedback. People review that analysis and decide how feedback informs the standard. The privacy notice explains how consultation responses are handled.

Factual claims and citations in public materials are checked against original sources. Any AI-generated visual or audiovisual material published by HPF will be disclosed.