Glossary
Plain explanations of the terms you will meet in AI video tools, plans and tutorials.
Terms are grouped by where you meet them. Each one is explained in a sentence or two, with a link to the page that covers it where there is one.
Generating clips
| Term | Meaning |
|---|---|
| Model | The AI that makes the video, such as Seedance 2.5, Kling 3.0 or Veo 3.1. One app can offer several models. See Models compared. |
| Prompt | The text instruction you give the model: subject, action, setting, camera, light and sound. See Prompting. |
| Negative prompt | A list of things you do not want. Few video models have a separate field for it; most work better if you describe what you do want. |
| Text to video | The model makes a clip from a prompt only. Good for exploring ideas and wide shots where nothing has to match. |
| Image to video | The model animates an image you give it, usually as the first frame. Use it for anything that must look exactly right, such as a character or product. |
| Reference to video | The model uses your images, videos or audio as references (for a face, an outfit, a motion, a voice) without copying them as the first frame. |
| First frame | The image the clip starts from. A good first frame means you can write a shorter prompt. |
| Last frame | The image the clip must end on. With a first and last frame, the model fills in the motion between them. Keep the two frames close. |
| Keyframe | A frame you fix at a point in the clip. Some models accept several, for example Luma Ray3.2 takes up to 16. |
| Multi shot | One generation that contains several shots with cuts, for example "Shot 1, Shot 2" in Seedance or Kling. |
| Extend | Continue an existing clip with a new generation. Veo 3.1 adds 7 seconds per extension. |
| Video edit | Change part of an existing clip (an object, a color, a background) and keep the rest. Also called video to video. |
| Draft mode | A cheap low resolution preview (480p on Seedance 2.5) that you re-render at full quality only if you like it. |
| Take | One generation of a shot. You usually make several takes and keep the best. Generating again is called a retake or reroll, and each one costs credits. |
| Seed | A number that controls the random start of a generation. Reusing it can bring back a similar result, but most closed models do not guarantee it. |
| Open weights | A model you can download and run on your own GPU, such as MiniMax H3 or LTX-2.5. |
| Moderation | The automatic filter that blocks prompts or references, for example real faces or blood. A blocked generation is a rejection. |
References and planning
| Term | Meaning |
|---|---|
| Reference image | An image you attach so the model knows what something looks like. Give each one a role: character, product, location or style. See Reference images. |
| Character sheet | One image showing a character from several angles on a plain background. Attach it to every shot to keep the face and outfit the same. |
| Location sheet | The same idea for a place: several views of one room or street, so the setting does not change between shots. |
| Storyboard | One image per shot, in order, that shows what the video will look like before you spend credits on it. See Storyboard. |
| Storyboard sheet | A whole storyboard in one image, for example a 3x3 grid. Some models can animate it panel by panel. |
| Shot list | The plan of the video as a list of shots with length, framing, action and camera move. See Script and shot list. |
| Elements | Kling's saved characters, built from 2 to 4 images or a short video, optionally with a voice. |
| Style line | One fixed sentence that describes the look. Reuse it word for word in every prompt to keep shots consistent. |
Shots and editing
| Term | Meaning |
|---|---|
| Shot | One continuous piece of video between two cuts. AI shots are usually 3 to 8 seconds. |
| Establishing shot | A wide shot at the start of a scene that shows where we are. |
| B-roll | Extra footage that supports the main story, such as close ups of hands, objects or places. Easy to generate and useful to hide cuts. |
| Aspect ratio | The shape of the frame: 16:9 for YouTube, 9:16 for TikTok, Reels and Shorts, 1:1 for square. Pick it before you generate. |
| Camera move | How the virtual camera moves: push in, pull out, pan, tilt, tracking shot, orbit. Use one move per shot. |
| Hook | The first 1 to 3 seconds that make someone keep watching. |
| Color match | Adjusting clips from different models so their colors look like one film, often with one LUT on all clips. |
Audio
| Term | Meaning |
|---|---|
| Native audio | Sound generated by the video model together with the picture: dialogue, effects, ambience, sometimes music. |
| Lip sync | Making a character's mouth match the speech. Done by the video model or by a separate tool such as Hedra or HeyGen. |
| Voiceover | A narrator's voice on top of the video, often made in ElevenLabs. |
| SFX | Sound effects: footsteps, doors, impacts, ambience. |
| Voice clone | A synthetic copy of a voice made from recordings. Use only your own voice or one you have permission for. |
Money and plans
| Term | Meaning |
|---|---|
| Credit | The unit apps charge in. A clip costs a number of credits that depends on model, length, resolution and audio. See Pricing. |
| Rollover | Unused credits moving to the next month. Most plans do not roll over. |
| Auto renew | The plan charges you again at the end of the period. First month discounts renew at full price. See Avoid traps. |
| Unlimited | A marketing word. On most platforms it means one model, for a few days, at capped resolution, in a slower queue. |
| Aggregator | A platform that sells access to many models from different makers under one plan, such as Higgsfield, Krea or OpenArt. See Platforms. |
| API | A way to call a model from code and pay per second. Used by developers and AI agents. |
| Cost per usable second | What you really pay for each second you keep, after all retakes. The best number for comparing tools. |
Finishing
| Term | Meaning |
|---|---|
| Resolution | The size of the picture: 480p and 720p for drafts, 1080p for most social video, 4K for big screens. |
| Upscale | Increasing resolution after generation, for example 720p to 4K with Topaz or a built in upscaler. Upscale only the takes you keep. |
| Frame interpolation | Adding frames between existing ones to make motion smoother, for example 24 to 30 fps. Can ghost on fast motion. |
| Watermark | A visible logo on the video, common on free plans. Many models also add an invisible one, such as Google's SynthID. |
| AI label | The tag platforms require on realistic AI content, such as YouTube's "Altered or synthetic content". See Finishing. |