Glossary of AI video production
Quotes, schedules and approval rounds are full of words that are obvious in our trade and nowhere else. Here are the ones that matter, explained briefly and the way we actually use them. Every term leads to a page of its own with examples, common questions and the person who owns the topic here.
Terms from A to Z
AI animation
Moving images whose frames are produced wholly or partly by a model. The difference to classic animation is not in the result but in the route: instead of building every movement by hand, we describe the scene and choose among many results.
AI avatar
A generated person speaking to camera. Common for product introductions, training and ads where somebody explains something without a shoot. An avatar is not a likeness of a real person, and that is exactly the difference to a deepfake.
AI film
A film whose images are generated rather than shot. Concept, direction, selection, editing and sound stay human work. AI film is therefore not a genre but a production route, and it suits brand films, commercials and social ads alike. What it costs is on pricing.
AI voice
A synthetically produced voice-over. It is quick, available in many languages and changeable without a studio booking. On long texts and on emotion the difference to a real voice artist is still audible. We say in advance when a real recording is worth it, and charge it by effort.
Approval point
A point in the project where you sign off before work continues. We use four: concept, styleframes, motion tests and rough cut. The purpose is to move changes to the place where they are still cheap. The approval points and what each one costs to change are set out in the offer.
Aspect ratio
The ratio of width to height. 16:9 is the landscape format for websites and YouTube, 9:16 the portrait one for Reels and Shorts, 1:1 the square for the feed, 4:5 the step in between that takes more room in the feed. We agree the formats before production because they decide the composition.
B-roll
Supporting shots cut in between the main ones: hands at work, a look out of the window, the product in detail. B-roll rarely carries the message, but it keeps the film moving and covers cuts. In AI production it is the part that works most reliably.
Character consistency
Keeping the same figure looking the same across several shots: same face, same clothes, same hair. This is the hardest technical problem in generative video. We solve it with reference images, fixed descriptions and, where needed, manual work. The longer a figure is on screen, the more effort it takes.
Colour correction
Matching brightness, contrast and colour across all the shots so the film feels like one piece. A deliberate colour treatment, called grading, usually follows and gives the film its mood. With generated shots this step matters even more, because every shot comes out of its own computation.
Deepfake
A convincing likeness of a real person, created or altered with AI. Since 2 August 2026, Article 50 of the AI Act requires disclosure for deepfakes. We do not create likenesses of real people without their consent. Whether a film has to be labelled depends on what it shows and where it runs, and we settle that before production.
Frame rate
The number of single images per second, written fps. Advertising usually runs at 25 fps, cinema at 24. Slow motion needs a higher capture rate, 50 or 100 fps, to stay smooth on playback. Generated shots often come out at 24 fps and are matched for the edit.
Generative AI
Models that produce new content instead of only sorting or recognising what exists. For video that means a shot appears from a description or an image where there was none. The models are trained on large amounts of data and do not necessarily return the same result twice.
Hook
The first one to three seconds of an ad, which decide whether someone keeps watching or scrolls on. A hook is not a title sequence, it drops the viewer into the middle of something: an unusual shot, a question, an image that irritates. For social ads we build several hooks for the same film so you can test them against each other.
Image-to-video
A method that turns a still image into a moving shot. The image fixes composition, light and figure, the model adds the motion. We work this way almost always, because the still can be approved beforehand. The counterpart is text-to-video, where the shot comes from a description alone.
Inpainting
Replacing one area of an image while the rest stays untouched. We use it to fix single errors, a wrong label or a hand with too many fingers, without generating the whole shot again.
Keyframe
An image that marks a fixed state, with the motion between keyframes calculated. In AI production we often set a first and a last image and let the model produce the way in between. That gives the movement a direction we can see in advance.
Lipsync
Mouth movement matching the sound. Short lines work well generatively, long dialogue in close-up stays demanding. When a figure speaks a lot, we suggest filming those passages and building the world around them with AI.
Moodboard
A collection of images, colours and references that sets a direction before anything of your own exists. It does not replace design, it settles expectations: two people saying "premium" often mean different things. A moodboard makes the difference visible while it still costs nothing.
Moving image
An umbrella term for anything that moves on screen: film, animation and generated video. Agencies use it when the technique should stay open. A moving-image budget says nothing yet about whether the work will be shot, animated or generated.
Portrait format 9:16
The upright format for Instagram Reels, TikTok and YouTube Shorts. A film built for 16:9 cannot simply be cropped: what sits at the edge disappears and text ends up in the wrong place. So we plan the formats beforehand and keep what matters near the centre.
Post-production
Everything after the shoot or the generation: editing, colour, sound, graphics, versions and delivery formats. In AI production the weight shifts here, because generating is fast and selecting, fixing and assembling is what takes the time.
Prompt
The description a model turns into an image or a shot: subject, camera, light, mood, movement. A negative prompt names what should not appear. Prompts are craft, not magic, and on their own they do not make a film. What makes the difference is the choice among a hundred results.
Rough cut
The first continuous version of the film, with order and length in place but colour, sound and detail still missing. The rough cut is our fourth approval point: it is the first time you can see whether the story holds.
Seed
A number that fixes the randomness in a model. The same prompt with the same seed gives a similar result, a different seed gives a variant. We keep a note of seeds so a shot that worked can be reproduced later, for a second language version for instance.
Showreel
A short cut of an agency's or a person's best work. It shows handwriting and range, not single projects in full. Ours runs on the home page, the individual films are on brand films.
Social ads
Paid advertising in social networks, usually short, usually portrait, usually in several versions for the same purpose. Unlike a brand film, a social ad is not produced once and shown for years, it is swapped constantly. That is why generative production pays off here in particular.
Sound design
Everything audible other than music and speech: footsteps, doors, wind, the click of a latch. Sound design creates a credibility nobody notices on purpose. With generated images it matters even more, because the shots arrive silent and feel artificial without it.
Spec work
Work made without a client, to show what is possible. Every film in our portfolio is spec work, some for invented brands, some as a free interpretation of known ones. We write that on each film, because a portfolio without the note pretends to be a client list it is not.
Storyboard
The film as a sequence of drawings or images, one per shot, with notes on camera, movement and sound. It settles before production whether the story holds. In AI production the storyboard often grows straight into styleframes, because we are generating the images anyway.
Styleframe
A still in the final visual language: look, light, colour and composition as they are meant to appear in the film. Styleframes are our second approval point and the most important one. If the look is wrong here, the change costs almost nothing. After production the same change costs a multiple.
Subtitles
Text on screen carrying what is said. In social networks most videos run muted, which makes subtitles a precondition rather than an extra. Burned-in subtitles play everywhere, switchable ones can be translated and searched.
Synthetic media
An umbrella term for image, sound and video produced wholly or partly by machine. It turns up above all in rules and platform policies, for example where YouTube or Meta require a label. Nobody says it in conversation, contracts and statutes do.
Teaser
A short version meant to make people curious without showing everything, usually five to fifteen seconds. One brand film yields several teasers for different channels. Because existing material is reassembled, a teaser costs far less than a film of its own.
Text-to-video
A shot made from a written description alone, with no reference image. It is fast and good for exploring at the start of a project. For finished production we mostly use image-to-video, because the result can then be approved in advance.
UGC video
A video in the style of a real user clip: handheld, natural light, room tone, no studio. UGC stands for user generated content. As an ad this style often works better than polish, because it does not look like advertising in the feed. Anyone advertising this way has to label the ad as such.
Upscaling
Computing an image or film up to a higher resolution. Generated shots often come out smaller than the final size, and upscaling brings them to 4K. It does not invent detail that was never there, but it gets considerably more out of what is.
Usage rights
What you may do with the finished film, where and for how long. With us the rights pass on payment in full and then cover your own digital channels worldwide and without a time limit, including paid advertising on them. Television, cinema and out-of-home are agreed separately. The detail is in our terms.
Frequently asked questions
Why does an agency need a glossary?
Because quotes and approval rounds are full of words nobody explains. Knowing what a styleframe is makes it obvious why approving at that point saves money. The glossary exists to make the conversation easier, not to make us look clever.
A term is missing, what now?
Tell us and we will add it. The list grew out of real questions and keeps growing the same way.
Do these definitions apply generally or only here?
The terms are general, the practice sections describe how we work with them. Where another agency does it differently, that is stated as our decision rather than as a rule.
Read on: Pricing • AI video or traditional shoot • What a commercial costs • About us
Missing a term?
Tell us which one. We will add it, and we explain it in the first call anyway.
Get in touch