GenAI Vocabulary 101
Fifteen terms, five minutes, zero jargon-panic. Learn the vocabulary that makes every AI tutorial, changelog, and Discord thread suddenly make sense.
Every craft has its handshake vocabulary. Chefs say *mise en place*, surfers say *point break*, and AI people say *inference* while nodding meaningfully. The good news: the core GenAI dictionary is small. Master the fifteen or so terms below and you can read any model announcement, follow any tutorial, and hold your own in any AI conversation. Let us get you fluent.
The big four
| Term | What it means |
|---|---|
| Prompt | The text instruction you give a model. The single highest-leverage skill in all of GenAI. |
| Model | The trained network that generates — Imagen 3, Sora 2, Suno are all models. |
| Token | The chunk of text a language model reads — about three-quarters of a word. |
| Inference | Running a trained model to generate output. Training happens once, in a lab; inference happens every time you press Generate. |
Image-speak
- Text-to-image — type a description, get a picture.
- Image-to-image — feed a picture in, get a transformed picture out.
- Inpainting — regenerate only a selected region of an image (fix that one weird detail).
- Upscaling — enlarge an image while adding real detail, not just stretching pixels.
- Aspect ratio — the canvas shape: 1:1, 16:9, 9:16.
- Seed — the random starting point; lock it to reproduce a result.
Each of these has a full entry in the Academy glossary — inpainting, upscaling, and aspect ratio are the ones you will meet first when editing images in VAR2. Bookmark-worthy honorable mention: the negative prompt, where you list what you *do not* want.
Words you will hear in the wild
- Multimodal — one model that handles several media types: text, images, audio, video.
- Parameters — the billions of learned numbers inside a model; roughly, its brain size.
- Open weights — model files anyone can download and run on their own hardware.
- Fine-tuning — extra training that specializes a general model for a specific job or style.
- Hallucination — confident output that is simply wrong. Charming in art, dangerous in facts.
Vocabulary in action
Isometric illustration of a tiny startup office at night, pastel color palette, clean vector style, soft ambient glow, 1:1 aspect ratio
One sentence, four vocab words doing real work: a style (isometric, vector), a palette (pastel), a mood (soft glow), and an aspect ratio (1:1). This is what fluency buys you.
Want to see how the pros wire these words together? Browse the model pages — start with Imagen 3 and Sora 2 — then head to Capabilities and Limits to learn what all this vocabulary can and cannot deliver.
See every one of these terms doing real work in ready-made, remixable examples. Browse templates