AI video has its own language — and half the confusion beginners feel is just vocabulary. This glossary defines 40+ essential terms, from generation basics to editing concepts, in plain language. Bookmark it and refer back as you learn.
Generation basics
Text-to-video (T2V): generating video from a written prompt alone. Maximum creative freedom, minimum control.
Image-to-video (I2V): animating a still image — the model generates motion from your picture. More control over composition and characters than T2V.
Video-to-video (V2V): restyling or transforming existing footage — e.g., turning a daytime clip into night, or live action into animation.
Prompt: the text description guiding generation. See prompt anatomy.
Negative prompt: text specifying what to exclude (artifacts, watermarks). See the negative prompt guide.
Seed: a number controlling the randomness of a generation. Same prompt + same seed = reproducible result (on most tools).
Reference image: an image supplied to anchor a character, object, or style across generations. See the reference workflow.
Models and tools
Diffusion model: the dominant architecture for image/video generation — iteratively denoises random noise into coherent output.
Transformer-based video model: newer architecture (e.g., Veo, Seedance) using attention mechanisms for better coherence over time.
Foundation model: a large general model (Veo 3, Kling 3.0) that platforms build features on top of.
API: programmatic access to a model (e.g., BytePlus ModelArk for Seedance) — for developers and bulk workflows.
Credits: the unit of paid usage. Generations consume credits based on duration, resolution, and features.
Shot and camera language
Establishing shot: wide shot setting the scene’s geography.
Medium shot: subject from roughly the waist up — the conversational default.
Close-up: face or detail filling the frame — emotion and emphasis.
Dolly / tracking: camera physically moving through space (in, out, or alongside).
Pan / tilt: camera rotating horizontally / vertically from a fixed position.
Orbit: camera circling the subject.
Aerial: high overhead shot, typically drone-style.
POV: point-of-view — camera as a character’s eyes.
Match on action: cutting mid-movement between shots for invisible transitions.
180-degree rule: keep the camera on one side of the action’s axis to preserve screen direction.
Quality and artifacts
Morphing: subjects or backgrounds warping mid-clip — the classic AI artifact.
Flicker: frame-to-frame brightness or texture instability.
Temporal consistency: how well elements persist coherently across frames — the core quality metric for AI video.
Character consistency: the same character remaining recognizable across shots. See the full guide.
Artifact: any unintended glitch — extra fingers, warped text, floating objects.
Upscaling: increasing resolution after generation (often AI-assisted).
Interpolation: generating intermediate frames to raise frame rate or smooth motion.
Production workflow
Storyboard: panel-by-panel visual plan of the video. See storyboard-first.
Shot list: the textual plan: every shot’s framing, camera, action, and lighting.
Animatic: a timed slideshow of storyboard panels — a rough draft of pacing.
Selects: the keeper clips chosen from all generations.
Rough cut: the first assembly — story order and pacing, no polish.
Fine cut: tightened edit with transitions and rhythm locked.
Picture lock: the edit is final; only sound and color remain.
Grade: the color styling pass. See the grading guide.
Mix: the final audio balance of dialogue, SFX, ambience, and music.
Master: the high-quality final export from which platform versions are made.
Publishing
Aspect ratio: frame proportions — 9:16 vertical (shorts), 16:9 widescreen (YouTube).
Bitrate: data per second of video — higher means better quality. See export settings.
Retention: the percentage of viewers still watching at each moment — the key metric for short-form.
Hook: the opening 1–3 seconds that decide whether viewers stay.
AI disclosure: labeling content as AI-generated where platforms require it (YouTube mandates it for realistic synthetic content).
Learn the terms, then learn the craft: start with the prompting system and work through the complete workflow.