A ↗

AI agent

Agentic AI

An AI system that can take steps towards a goal, such as finding information, using a tool or checking its own work, with rules about what it may do.

A ↗

AI animation

Generative animation

Animation created or assisted by AI tools. The tools may generate movement, backgrounds or characters, while people still direct the story, look and final edit.

A ↗

AI artefact

Generation artefact, visual artefact

An unwanted flaw introduced by an AI tool, such as flickering detail, distorted hands, unstable lettering or a strange sound. An artefact may be visible even when the scene itself makes sense.

A ↗

AI avatar

Virtual presenter

A digital on-screen person whose appearance or performance is generated with AI. An avatar may be fictional or based on a real person with their permission.

A ↗

AI disclosure

AI transparency

A statement that AI was used to create or substantially alter content. The appropriate wording and placement depend on the audience, client requirements, platform rules and the production itself.

A ↗

AI model

The underlying system that processes an input and produces an output such as text, images, audio or video. Models differ in what they can do, how much control they offer and what use their providers permit.

A ↗

AI video production

AI-assisted video production

Making video with AI somewhere in the production process, from concept images and storyboards to generated footage, editing, sound or localisation.

A ↗

Animatic

A timed version of a storyboard, usually combining still images with temporary dialogue, music or sound. It lets a team assess pacing and storytelling before producing final shots.

A ↗

Aspect ratio

16:9, 9:16, 1:1

The shape of a video frame, expressed as its width compared with its height. Different channels need different shapes, such as wide video or vertical social content.

B ↗

B-roll

Supplementary footage

Supporting shots used alongside the main action or interview. AI can help create some B-roll when the scene is difficult to film or is clearly illustrative.

B ↗

Brand safety

Brand governance

The checks that keep AI-assisted content accurate, appropriate and consistent with a brand's identity, permissions and audience expectations.

C ↗

Character consistency

Character continuity

Keeping a character recognisably the same across shots, including their face, clothing, proportions and manner of moving.

C ↗

Character sheet

Character reference sheet

A set of approved views and details defining how a character should look. It can include different angles, expressions, clothing, proportions and other information needed to keep later shots consistent.

C ↗

Colour grade

Grading, colour grading

The finishing process used to adjust and balance colour across a film and create a consistent visual treatment. Generated and filmed shots often need grading so they sit comfortably together.

C ↗

Compositing

Visual effects compositing

Combining visual elements into one finished shot. Those elements might include filmed footage, animation, graphics and AI-generated backgrounds.

C ↗

Confidential inputs

Private inputs, sensitive assets

Client information or assets supplied to an AI system that are not intended for public use, such as unreleased products, scripts or internal documents. Their storage and processing may require additional safeguards.

C ↗

Content credentials

C2PA credentials

Information attached to a media file that records parts of its origin and editing history. It can help viewers and teams understand how an image or video was made.

C ↗

Copyright and IP ownership

IP rights, copyright ownership

The rights associated with creative material, including who may use, adapt or distribute it. In AI-assisted production, supplied assets, music, generated elements and the finished film may each raise different ownership or licensing questions.

C ↗

Creative brief

Production brief

The agreed starting document for a production, setting out the audience, objectives, message, deliverables, brand requirements, timings and other constraints that should guide the work.

D ↗

Deliverables

Delivery assets

The files or versions a production has agreed to supply, such as a master film, social edits, vertical versions, language versions, captions or still images.

D ↗

Delivery specification

Technical delivery specification

The technical requirements a finished file must meet, such as resolution, aspect ratio, frame rate, codec, audio configuration and file format. Broadcasters and platforms may each require different specifications.

D ↗

Diffusion model

Diffusion AI

A type of generative model that learns to turn noise into a coherent image or sequence, guided by instructions or reference material.

D ↗

Digital double

Digital replica

A digital likeness of a real person used in a production. Its use should be agreed with that person, including where, how and for how long it may appear.

F ↗

Fine-tuning

Model fine-tuning

Further training an existing AI model on selected examples so it behaves more consistently for a particular task or style.

F ↗

Frame interpolation

AI slow motion

Creating new frames between existing video frames to make motion smoother or slow footage down. The added frames are predictions, so they need checking for visual errors.

G ↗

Generative AI

Gen AI

AI that creates new material, such as text, images, sound or video, in response to an input. The output still needs human selection and review.

G ↗

Generative fill

AI fill

Using AI to create image content inside a selected area, usually to add, remove or replace something while matching the surroundings.

G ↗

Grounding

Grounded AI

Connecting an AI response to specific source material or facts instead of relying only on what the model has learned during training.

H ↗

Hallucination

AI fabrication

An AI output that presents invented or incorrect information as if it were real. In a production, this might be a false claim in a script or a convincing-looking detail that was never in the approved source.

H ↗

Human in the loop

Human review

A workflow in which people review, guide or approve AI outputs at important points. It keeps creative and factual decisions with the production team.

H ↗

Hybrid production

AI-assisted production

A production approach that combines AI-generated material with techniques such as live action, traditional animation, CGI, motion graphics, compositing or conventional post-production.

I ↗

Image-to-image

I2I

Creating a new image using another image as a starting point or guide. It can change style, setting, composition or details while retaining selected characteristics of the source.

I ↗

Image-to-video

I2V

Generating movement from a still image. The starting image provides a visual anchor while the model creates motion and, sometimes, camera movement.

I ↗

Inference

Model inference

The stage when a trained AI model produces an output from a new input. In production, each generated image, voice clip or video is an inference run.

I ↗

Inpainting

AI image repair

Regenerating a selected part of an image while leaving the rest in place. It can repair mistakes or replace an element within a shot.

K ↗

Key visual

Campaign key visual

A principal campaign image that establishes the approved look or central creative idea. It can become an important reference when producing related shots, formats and campaign assets.

K ↗

Keyframe

Animation keyframe

A frame marking an important point in an animation or edit, such as a change in position, scale or appearance. Some AI video tools also use selected frames as visual guidance for a generated shot.

L ↗

Lip sync

AI lip-sync

Matching a speaker's mouth movement to recorded or generated speech. AI can help adapt that movement when dialogue changes or a video is dubbed.

L ↗

Localisation

Video localisation

Adapting a video for another market or language. This can involve translation, voice, on-screen text, cultural details and different delivery formats.

L ↗

LoRA

Low-rank adaptation

A compact way to adapt an existing AI model to a particular subject or style without retraining the whole model.

M ↗

Model Context Protocol (MCP)

MCP

An open standard that lets AI applications connect to external tools and information sources through a common interface.

M ↗

Motion capture

Mocap, performance capture

Recording a person's movement so it can drive or inform a digital performance. AI-based techniques can increasingly extract useful body, face or hand motion from ordinary video.

M ↗

Multimodal AI

Multimodal model

AI that can work with more than one type of information, such as words, images, sound and video.

O ↗

Outpainting

Canvas extension

Generating new image content beyond the original frame. It can make room for a different crop or reveal more of a scene.

P ↗

Picture lock

Locked picture

The point at which the timing and order of the edit are approved and no further picture changes are expected. Sound, colour and other finishing work can then proceed against a stable cut.

P ↗

Previsualisation

Previz, pre-vis

A rough visual version of a film made before final production. It helps the team agree on shots, pacing and creative direction early.

P ↗

Product consistency

Product accuracy

Keeping the appearance of a product accurate across generated material, including its shape, colours, branding, packaging and important design details.

P ↗

Prompt

AI instruction

The instruction or material given to an AI model to guide its output. A prompt can include text, images, audio or other references.

P ↗

Prompt iteration

Prompt refinement

The process of adjusting instructions, references or settings over several generations to move an AI output towards the required result. It is usually a process of testing and selection rather than one perfect prompt.

P ↗

Proof of concept (POC)

POC, feasibility test

A limited test made to establish whether a proposed creative or technical approach can work before committing to full production. It normally concentrates on the most uncertain part of the idea.

P ↗

Provenance

Media provenance

The record of where media came from and how it was changed. For AI video, this may include source assets, tools, permissions and editing steps.

Q ↗

Quality control (QC)

QC, final checks

The final checking process used to find creative or technical problems before delivery. For AI-assisted video this may include continuity errors, incorrect branding, audio faults, captions and generated artefacts.

R ↗

Reference image

Visual reference

An image supplied to guide an AI output's subject, composition or style. It gives the model more concrete direction than words alone.

R ↗

Reference video

Video reference

Existing footage supplied to guide aspects of a generated result, such as movement, performance, timing, camera behaviour or composition. The amount retained depends on the technique and model used.

R ↗

Reproducibility

Repeatability

The ability to recreate substantially the same AI result using the same inputs and settings. Exact reproduction is not always possible because models, services and generation processes can change.

R ↗

Retrieval-augmented generation (RAG)

RAG

A method that retrieves relevant information from a chosen source before an AI model answers. It can make responses more useful for a specific project.

R ↗

Review round

Amends, revision round

A defined stage in which a client or internal team reviews work and requests changes. Agreeing how many rounds are included helps keep feedback organised and production scope clear.

R ↗

Rights and consent

Talent consent, usage rights

The permission and usage limits agreed for a person’s likeness, voice or performance, or for other material used in a production. The allowed channels, territories and duration should be clear before AI-assisted reuse.

R ↗

Rotoscoping

Roto, subject masking

Separating a person or object from its background across video frames. AI can speed up this detailed editing task, though edges often need checking.

R ↗

Rough cut

First cut, offline edit

An early edited version of a video used to assess structure, timing and content before detailed finishing. Shots, graphics, sound or AI-generated material may still be temporary.

S ↗

Scope of work (SOW)

SOW, project scope

The agreed description of what a production will include, such as deliverables, formats, languages, production stages, responsibilities and revision allowances. It helps distinguish planned work from later additions.

S ↗

Seed

Random seed

A number used to initialise part of a generation process. Reusing it can help reproduce or compare outputs, though other settings and model changes also matter.

S ↗

Shot continuity

Scene continuity

Keeping details consistent between shots so a sequence feels like one scene. This includes props, lighting, wardrobe, screen direction and timing.

S ↗

Shot extension

Video extension

Generating additional footage before or after an existing video segment so a shot lasts longer or continues beyond its original ending. New material must be checked for continuity with the source.

S ↗

Shot list

A structured list of the shots required for a video, normally describing their subject, action, framing and purpose. It helps turn an approved concept into a manageable production plan.

S ↗

Sign-off

Sign-off, approval gate

Confirmation that a particular production stage or asset has been accepted and work can proceed. Sign-off may apply to scripts, storyboards, style frames, shots, edits or final delivery.

S ↗

Sound design

Audio design

The creation and arrangement of sound effects, ambience and other audio elements that support the story and make images feel convincing. AI-generated visuals often still require substantial conventional sound work.

S ↗

Source files

Project files, editable files

The editable production files used to create a finished piece, rather than only the exported video. Whether these files are included in delivery should be agreed as part of the project scope.

S ↗

Speech-to-text

Transcription, STT

Turning spoken audio into written text. It supports transcripts, captions, searchable interviews and faster editing.

S ↗

Start and end frames

First frame, last frame, start/end frame

Still images supplied to guide the beginning and ending of a generated video shot. The model creates movement between them, although the transition and intermediate details still need review.

S ↗

Storyboard

Shot board

A series of images that maps out a video's main shots before production. AI can help produce early options, which the team then selects and refines.

S ↗

Style frame

Look frame

A polished still image showing the intended visual treatment of a film or sequence. Style frames help agree composition, colour, lighting and design before multiple shots are produced.

S ↗

Style transfer

Visual style transfer

Applying characteristics of one visual treatment to different imagery while trying to preserve its underlying subject or movement. Results vary depending on how strongly the source structure is retained.

S ↗

Synthetic media

AI-generated media

Images, sound or video created or substantially altered by computer generation. The term covers many uses, from fictional scenes to realistic voices.

T ↗

Temporal consistency

Frame-to-frame consistency

The stability of details from frame to frame within a generated video. Poor temporal consistency can cause faces, objects, textures or backgrounds to change unexpectedly while a shot plays.

T ↗

Text-to-image

T2I

Creating a new image from a written description. It is commonly used for visual development, storyboards, style exploration and creating starting frames for AI video.

T ↗

Text-to-speech

TTS, synthetic voice

Generating spoken audio from written text. The voice may be generic or, with permission, modelled on a particular person.

T ↗

Text-to-video

T2V

Generating moving images from a written description. The result may be useful for exploration, but detailed creative control usually takes further work.

T ↗

Training data

Model training material

The examples used to teach an AI model patterns before it is used. Their origin and permissions can matter when choosing tools for commercial work.

T ↗

Training opt-out

No-training setting, training exclusion

A provider setting or contractual commitment that submitted material will not be used to train or improve its general AI models. Its scope varies by service, account type and agreement, so the current terms should be checked.

T ↗

Treatment

Creative treatment

A document that explains the proposed creative approach before production begins. It may cover the story, visual direction, tone, characters, filmmaking style and examples of how the finished video could feel.

U ↗

Upscaling

Super-resolution

Increasing an image or video's resolution. AI tools may predict new detail, so the result needs checking rather than assuming it exactly matches the original.

V ↗

Video-to-video

V2V, video restyling

Using an existing video as a guide while AI changes its appearance. The original movement often remains the starting point.

V ↗

Voice cloning

AI voice replica

Creating a synthetic voice that resembles a particular person. It requires clear permission and careful limits on how the voice may be used.