Glossary: AI Video and Short Drama Terms

Short answer

This glossary defines forty terms across four areas: rights and likeness, disclosure and compliance, production craft, and platform mechanics. Each definition comes from the document that actually establishes it — statute text, regulation text, platform policy or product documentation — rather than from how the term gets used in conversation. Where a term is contested or unsettled, the entry says so.

Most glossaries in this category define terms by paraphrasing other glossaries. These are taken from the source that gives each term its meaning, which occasionally makes them narrower than you'd expect.

Legal terms below are defined from the documents that establish them, including one bill that hasn't passed. Definitions are not legal advice and several are being actively contested. Where a term's meaning isn't settled, the entry says so.

Rights and likeness

Digital replica: Under the NO FAKES Act as drafted, a newly created, computer-generated, highly realistic electronic representation readily identifiable as the voice or visual likeness of an individual. It also covers real recordings of someone where the fundamental character of the performance has been materially altered. The Act is a bill and has not passed.
Readily identifiable: The operative test in the digital replica definition. It asks whether a viewer could identify who a representation is meant to be — not how it was made, and not what the character was called.
Fully synthetic character: A character with no real-world counterpart. Nothing is readily identifiable as an individual, because no individual is being represented, so likeness obligations have nothing to attach to.
Right of publicity: A state-law right controlling commercial use of a person's name, image or likeness. It applies regardless of union status and varies substantially between states, including on whether it survives death.
Post-mortem right: A likeness right surviving death. Under the NO FAKES Act as drafted it would run ten years from death, renewable in five-year periods only where the right holder demonstrates active authorised commercial use in the preceding two years, with a hard ceiling of seventy years. Today it's a state question and varies.
Digital fingerprint: In the NO FAKES Act, an electronic identifier created by a cryptographic hash function unique to a specific piece of material. The bill's takedown process would use it to require removal of all publicly available copies matching a flagged replica, not just the one named in a notice.

Terms specific to Hexcoded's actor library

AI actor: A platform-provided or user-generated persona depicting nobody real. Generated from a text description, so no consent or payment attaches to it.
Human creator: A real person who submitted their own likeness under a licence. On Hexcoded they pass a liveness check and a face-match before their likeness can be published, their consent is a timestamped record, and they earn every time a paying brand generates with them.
Liveness check: A verification step confirming a submitted likeness belongs to a live person rather than a photograph or another person's image, processed through a third-party service. Publication is conditional on passing it.
Consent-verified receipt: Confirmation shown at casting that the creator you selected is a real, verified person who accepted the licence. Evidence rather than an assurance.
Faceless mode: Rendering with no on-screen presenter — voiceover and product visuals only. Available in Talking Actors, Make a Custom Video and URL → Ad.
Withdrawal: A creator's ability to stop new use of their likeness at any time. Content already generated and delivered stays valid; new generation stops.

Disclosure and compliance

Synthetic media: Realistic content generated or meaningfully altered by AI. The working term across YouTube, TikTok and Meta for what triggers a disclosure requirement.
Deepfake: Under Article 3(60) of the EU AI Act, AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events **and** would falsely appear authentic or truthful. Both parts must be met, so clearly fantastical content falls outside it.
Deployer: Under the EU AI Act, a person or body using an AI system under their authority, excluding personal non-professional use. Regular economic benefit, or a business, trade, occupational or freelance activity, makes a natural person a deployer.
Article 50: The transparency layer of the EU AI Act, applicable since 2 August 2026. Risk-independent — it applies whatever tier a system sits in. Two of its four routes fall on deployers: deepfake disclosure, and disclosure of AI-generated text published on matters of public interest.
Synthetic performer: Under New York's advertising disclosure law, a digitally created asset created, reproduced or modified by computer using generative AI or a software algorithm, intended to create the impression it is engaging in an audiovisual or visual performance. Currently the only US statutory term of its kind in advertising.
C2PA Content Credentials: A cross-industry metadata standard recording whether AI was involved in creating or editing a file. Platforms including TikTok and Meta read these credentials to detect AI content even without creator disclosure. They can be stripped by re-encoding.
Endorser: Under the FTC's endorsement guides at 16 CFR Part 255, a party whose opinions a message appears to reflect — and, since the 2023 revision, a party who **may appear to be** an individual, group or institution. That language is read as reaching virtual influencers and synthetic personas.
Clear and conspicuous: The FTC standard for a disclosure: difficult to miss, and easily understandable by ordinary consumers. A buried hashtag or a pinned comment fails the first part.

Terms about YouTube monetization

Inauthentic content: YouTube's term for mass-produced or repetitive video, whatever tool produced it. Renamed from "repetitious content" in July 2025. In July 2026 it was broken into three named categories without the underlying policy changing.
Content farming: YouTube's own description of the behaviour the inauthentic content policy targets — producing many very similar videos quickly, generic and without a narrative arc.
Community Guidelines: YouTube's baseline for removal — egregious harm, hate speech, nudity. A lower bar than monetization eligibility, and a separate system from it.
YPP policies: YouTube Partner Program policies, applying a higher bar than Community Guidelines and evaluating a channel as a whole rather than individual videos.

Production craft

Coverage: The set of shots gathered so a scene can be cut together — the establishing wide, singles on each character, over-shoulders, inserts. Coverage has continuity rules that individual shots don't.
Eyeline match: The convention that characters looking at each other appear to meet each other's gaze across a cut. In generated footage it has to be specified in world terms — toward the doorway — rather than frame terms, because frame terms flip on a reverse angle.
Screen direction: The convention that a character on the left stays on the left, and movement continues in the same direction across a cut.
The 180-degree rule: Keeping the camera on one side of an imaginary line between characters, so the geography of a scene holds across cuts.
Re-anchoring: What the first ten seconds of a short drama episode do instead of establishing. A returning viewer doesn't need the premise introduced; they need reminding where they were.
Forward move: The single plot development an episode delivers — a revelation, a confrontation or a decision. Short drama convention is exactly one per episode.
Suspension: Ending an episode unresolved. The convention is that resolution is what makes a viewer stop watching.
Negative prompt: Instructions describing what should not appear. The standard set for series work covers face swaps, wardrobe changes, broken hands, floating props and camera jumps.

Platform and model mechanics

Handle: The identifier on a saved reference, written as @name and callable in any prompt in any tool. Calling a handle rather than re-describing a character is what prevents identity drift across a series.
Reference pool: The assets attached to a single generation context. In Hexcoded's video engine, up to 50 total — up to 30 images, 30 actors, 10 videos and 10 audio. Image editing has a separate ceiling of 16. Per-model ceilings vary within that.
Create, Edit, Extend, Motion: The four video sub-modes in Creative Studio. Create generates from a prompt or keyframes; Edit modifies a clip of three to ten seconds; Extend appends to a rendered clip; Motion transfers movement from a driving video onto a character. The model list filters per mode.
Latent re-encoding: The mechanism by which small variations compound across chained generations. Hexcoded's model documentation names it as the cause of drift when subject descriptors are altered between Extend rounds — which is why restating descriptors identically matters.
Native audio: Audio generated in the same pass as the picture rather than synced afterwards. Seedance 2.5 processes visual frames and acoustic signals together in one generation. Availability differs per model: some offer a toggle, some have audio always on.
Keyframing: Anchoring a generation with a start frame and optionally an end frame, with the model synthesising the motion, lighting shifts and camera path between them.
Plan First: A free storyboard and script preview available in Make a Custom Video, Talking Actors and URL → Ad. It consumes no credits, which makes the planning stage free.
Light edits: The editing scope permitted where a licensed actor appears — trimming, cropping, captions and subtitles, music, end-cards. Re-generating, morphing or re-voicing a face outside the platform is not permitted and is listed as a prohibited use.

Copyright

Human authorship: The US requirement for copyright protection. Purely AI-generated material isn't copyrightable because there's no human author.
Selection and arrangement: One route to authorship over work containing AI material. Arranging AI-generated elements with human-authored content can be protectable as a compilation, though the AI elements aren't protected standing alone.
Iterative modification: The other route. Modifying AI output using editing tools in an iterative fashion can establish authorship, with a protection scope analogous to a derivative work.
De minimis: The threshold above which AI-generated content must be disclosed on a US copyright registration, along with a brief explanation of the human author's contribution.

Which of these terms aren't settled?

Four, and it's worth knowing which ground is contested.

Digital replica. The definition above is from a bill that hasn't passed and has already been revised once. Its final wording may differ.

Readily identifiable. No published threshold exists for how close a resemblance has to be. It's the question most likely to be litigated.

Deepfake. The EU definition is in force, but what counts as "clearly fantastical" and therefore outside it is being worked out through Commission guidance rather than settled.

Prompt authorship. The Copyright Office's position is that prompts alone don't establish authorship given current generally available technology — and it says explicitly that this could change as technology evolves.

Definitions of legal terms are taken from the documents that establish them, current as of the publication date. The NO FAKES Act is pending legislation and had not passed at the time of writing. Likeness law varies by state. Nothing here is legal advice.

The bottom line

"Readily identifiable" is the phrase that decides likeness questions. Learn it before anything else
A deepfake under EU law must both resemble something real and falsely appear authentic. Both parts
Deployer versus personal use is the EU scope test. It isn't geographic
"Clear and conspicuous" has a definition: difficult to miss, and easily understandable
Coverage, eyeline match and screen direction are century-old film craft. They apply unchanged to generated footage
A handle is the mechanism that prevents identity drift. Re-describing a character is what causes it
Light edits means trimming, cropping, captions, music and end-cards. Nothing touching performance
Purely AI-generated material isn't copyrightable. Your script, structure and edit are the protected layer

Under the NO FAKES Act as drafted, a newly created, computer-generated, highly realistic representation readily identifiable as the voice or visual likeness of an individual. It also covers real recordings materially altered in the fundamental character of the performance. The Act is a bill and hasn't passed.

Synthetic media is the broad platform term for realistic AI-generated or meaningfully altered content. A deepfake is narrower — under EU law it must resemble existing persons, objects, places, entities or events and would falsely appear authentic. Content that's obviously not authentic isn't a deepfake.

It asks whether a viewer could identify who a representation is meant to be. It doesn't ask how the representation was made or what the character was called, which is why a fictional character generated from a real person's photographs still counts.

The set of shots gathered so a scene can be cut together — establishing wide, singles, over-shoulders, inserts. Coverage carries continuity rules that individual shots don't, because an audience reads a cut as continuous space and time.

The identifier on a saved reference, written as @name and callable in any prompt in any tool. Calling a handle rather than re-describing a character is the mechanism that keeps a face consistent across a series.

Four. Digital replica and its "readily identifiable" test both come from a bill that hasn't passed and has been revised once. What counts as "clearly fantastical" under the EU deepfake definition is still being worked out through guidance. And the Copyright Office says its position on prompt authorship reflects current technology and could change.

The vocabulary, in the product

Handles, reference pools, the four video modes, native audio toggles. Creative Studio runs 30+ models on one credit balance, with every setting visible before you generate.

Open Creative Studio

More on workflow, continuity and shot planning in Production.