Creative AI vocabulary
Make with AI tools. 28 hand-written terms with real workplace examples.
Learn creative ai terms every day.
One hand-written term. A real workplace example. Thirty seconds before your day gets loud.
Text-to-Image
Creating a new still image from a written prompt. You describe the subject, setting, light and framing, and the model renders a picture from that description, no camera or source photo needed.
Style Reference
An image you give the model to borrow a look from: palette, light, texture and mood. The prompt still decides what is in the picture; the reference decides how it feels.
Inpainting
Regenerating only a masked area of an image, such as a stray cable or a wrong label, while the rest of the frame stays as it was. You mask the spot, describe the fix, and the model blends it in.
Upscaling
Raising an image's resolution so it holds up at larger sizes, with the model inventing plausible fine detail as it enlarges. That invented detail can go wrong, so check faces, small text and logos.
Text-to-Video
Generating a short video clip from a written prompt. You describe the scene, the action and the camera move, and the model renders moving footage with no camera, set or source image.
Image-to-Video
Animating a still image into a short clip. The still usually becomes the first frame and fixes the look and composition, while the prompt describes the motion, so you keep more control than with text alone.
Character Consistency
Keeping a person or mascot recognizably the same in every shot: face, hair, outfit, proportions and markings. Generators drift from one generation to the next, so teams hold the look with reference images.
Lip Sync
Matching a speaker's mouth movements to an audio track. AI lip sync reshapes the mouth in existing or generated footage so a new line, translation or voiceover looks spoken on camera.
Seed
The number that starts a generation's randomness. Reuse it with the same prompt, model and settings to get the same result back; change it and you get a fresh variation.
Aspect Ratio
The shape of a frame, written as width to height: 16:9 widescreen, 9:16 vertical, 1:1 square, 4:5 portrait. Set it before you generate, since cropping later cuts into the composition.
Negative Prompt
A list of things to keep out of a generation, such as text, watermarks, extra fingers or clutter, kept separate from the main prompt. Some tools have a field for it; others take exclusions in plain words.
Prompt Adherence
How closely an output follows what the prompt asked for: the right subject, count, colors, positions and words. Weak adherence looks like a red scarf that comes back blue or a third person who never shows up.
Voice Clone
A synthetic copy of a real person's voice, built from recordings of them, that can say lines they never recorded. Make one only with that person's clear consent and agreed terms of use.
Text-to-Speech
Turning written text into spoken audio with a synthetic voice. Teams pick a stock or licensed voice, set the pace and emphasis, and use it for scratch reads, explainers and accessible versions of content.
Likeness Rights
A person's legal right to control commercial use of their face, voice or likeness, often called the right of publicity. Cloning or generating a recognizable person for an ad needs their consent.
Model Release
A signed consent form from a person who appears in a photo, video or ad, allowing use of their image for agreed purposes. Here model means the person in front of the camera, not an AI model.
Creative Testing
Running ad variations against each other with real audiences to learn which idea, hook or visual performs best. Change one thing at a time so you know what made the difference.
Dynamic Creative
An ad setup where you upload headlines, images, videos and calls to action, and the platform assembles the combinations automatically, then shifts delivery toward the ones that perform.
Creative Fatigue
The drop in an ad's performance as the same audience sees it too often. Click-through slips, costs per result climb, and the usual fix is fresh creative rather than more budget.
ROAS
Return on ad spend: revenue from ads divided by what the ads cost, so $40,000 in sales on $10,000 of spend is a ROAS of 4. It ignores margins, so a healthy-looking ROAS can still lose money.
Synthetic Media
Images, video, audio or text generated or substantially altered by AI, rather than shot, recorded or written by people. The umbrella covers a generated product shot, a cloned voice and a face swap alike.
Content Credentials
Tamper-evident metadata attached to an image, video or audio file that records how it was made and edited, including any AI use. It is the consumer name for the C2PA open standard.
Brand Safety
Keeping a brand's ads away from harmful or unsuitable content and placements, such as violent news or hate speech. With generated work, teams hold what the ad itself shows to the same standard.
Rights Clearance
Confirming you have permission to use everything in a piece before it runs: music, footage, artwork, fonts, locations and people. With AI work, that includes the reference images and source files you fed in.
Storyboard
A sequence of frames sketching each shot of a spot in order, with notes on action, dialogue and camera, so everyone agrees on the story before anything is shot or generated.
Mood Board
A collage of images, colors, type, textures and references that sets the look and feeling of a project before production starts. It aligns the team on tone, not on specific shots.
Previs
Short for pre-visualization: rough versions of shots made before the real shoot to test framing, timing and camera moves. Teams now often generate previs with AI video tools.
Asset Library
The organized, searchable home for a brand's approved files: logos, product shots, footage, fonts, voices and generated assets, each tagged with its source and usage rights so nobody grabs a stale or unlicensed file.
Building toward a role? Browse career pathways that map terms to jobs.