The AI Glossary for Event and Exhibition Designers

Forty terms, from diffusion models to digital twins, explained for people who design spaces.

This glossary explains forty AI terms you will hear in software demos, from diffusion models to digital twins. Each term gets a plain definition and a line on why it matters when you design rooms and stands. It ends with ten questions that help you test any tool's claims.

This AI glossary for designers defines forty terms from software demos for events, exhibitions and interiors. Words like render, model, twin and scale get used loosely. The gap between what a tool claims and what it does often hides in one of them. Each entry gives a plain definition and one line on why it matters to people who design rooms and stands.

How to use this AI glossary

The forty terms sit in five tables of eight, in the order work happens. Related terms share a row, so DXF sits under DWG. Two distinctions matter most. A raster image has no scale, and a vector drawing does. A digital twin stays in step with the real place, and a 3D tour does not. For the wider picture, read what AI floor plan generation is.

How the models work

• Model. What it means: A system that learned its behavior from examples, not written rules. For designers: Ask whether it outputs text, pictures or measured geometry.
• Training data. What it means: The examples a model learned from. Gaps become blind spots. For designers: A model trained on homes may misjudge a hall.
• Large language model. What it means: A text model built on the transformer (Vaswani et al., 2017) that predicts the next token. For designers: Strong on briefs, rules and sums; blind to geometry it is not given.
• Token. What it means: The unit a model reads and writes, often part of a word. For designers: Limits and prices are counted in tokens.
• Context window. What it means: How much a model can consider at once. For designers: A long exhibitor manual may not fit.
• Multimodal. What it means: Takes several kinds of input, such as text, images and PDFs. For designers: Reading a plan image is not measuring it.
• Diffusion model. What it means: Makes images by removing noise step by step (Ho et al., 2020). Latent diffusion, in compressed form, made it fast (Rombach et al., 2022). For designers: Pictures of rooms, not measured rooms.
• Hallucination. What it means: Fluent, confident output that is wrong. For designers: A confident wrong exit width is one. Check it against the drawing.

How images are made

• Prompt. What it means: The instruction given to a model. For designers: Dimensions in a prompt are requests, not guarantees.
• Negative prompt. What it means: What the image should leave out. For designers: Removes clutter such as stray chairs.
• Seed. What it means: The number that fixes a generation's random start. For designers: Record it to reproduce a view.
• Image-to-image. What it means: A new image made from an input image plus a prompt. For designers: Keeps a stand's composition while testing finishes.
• Inpainting. What it means: Regenerating one masked area and keeping the rest. For designers: Changes a counter without redrawing the stand.
• Outpainting. What it means: Extending an image past its edges, also sold as generative expand. For designers: The added space was never designed.
• Style reference. What it means: An image whose look the model copies. For designers: Keeps one stand consistent across views.
• Upscaling. What it means: Enlarging an image with predicted pixels. For designers: It can invent detail nobody designed.

How tools are steered

• Fine-tuning. What it means: Further training on a smaller, specific dataset. For designers: Often what "learns your studio's style" means. Ask whose data trains it.
• LoRA. What it means: A small add-on trained on top of a model (Hu et al., 2021). For designers: Cheap style training. Ask where your images go.
• Embedding. What it means: Numbers that place items by meaning, so similar ones sit close. For designers: Powers search across past projects.
• Retrieval-augmented generation. What it means: Searching your documents first, then answering from them. For designers: Answers are only as current as the documents.
• Agent. What it means: A model that plans steps and uses tools to finish a task. For designers: Ask which actions it takes without approval.
• Model Context Protocol. What it means: An open standard for connecting models to tools and data, used by Autodesk Assistant in Revit 2027 and AutoCAD 2027. For designers: The assistant can act on your live model.
• Inference. What it means: Running a trained model to get an output. For designers: Where it runs decides where your files go.
• Temperature. What it means: How random the output is. For designers: Higher for varied concepts, lower for repeatable answers.

How space is drawn and captured

• Vector drawing. What it means: Lines and shapes stored as coordinates in known units. For designers: Can be measured, scaled and edited.
• Raster image. What it means: A grid of pixels with no units. For designers: Cannot be measured reliably.
• Scale. What it means: The ratio between drawing and reality, such as 1:200. For designers: A plan without units or scale is a picture.
• DWG. What it means: AutoCAD's native format. DXF is its exchange cousin. For designers: The file venues and crews ask for most.
• BIM. What it means: Building information modeling: 3D objects that carry data such as size and material. For designers: A venue's BIM model holds exits and heights you can query.
• IFC. What it means: buildingSMART's open standard for exchanging BIM data. For designers: Shares a model across different software.
• glTF and GLB. What it means: The Khronos Group's 3D scene format. GLB is the single-file version. For designers: Opens in browsers and real-time viewers.
• Point cloud. What it means: Millions of measured points from a scanner or photos. For designers: The raw survey behind a twin, not yet a drawing.

How plans are generated and tested

• Photogrammetry. What it means: 3D geometry built from overlapping photographs. For designers: Accuracy depends on photos and reference measurements.
• Gaussian splatting. What it means: A captured scene stored as soft 3D blobs that render in real time (Kerbl et al., 2023). For designers: Looks photographic; it is not a measured model.
• Generative design. What it means: Software that produces many options from goals and constraints. For designers: Options are only as good as the rules.
• Constraint. What it means: A rule the output must meet, such as a 2 m clear path to every exit. For designers: Ask if it is enforced or only requested.
• Test fit. What it means: A quick layout proving a program fits a space. For designers: The first check that a brief fits a room.
• Crowd simulation. What it means: Software that models people moving through a layout. For designers: Shows pinch points before the doors open.
• Digital twin. What it means: "A virtual representation of real-world entities and processes, synchronized at a specified frequency and fidelity" (Digital Twin Consortium, 2020). For designers: A 3D tour is not a twin unless it stays in step with the building. More in what a venue digital twin is.
• Real-time rendering. What it means: Images drawn instantly as the viewer moves. For designers: Good for walkthroughs; ask how lighting is calculated.

Ten questions to ask in any AI software demo

1. Is the output vector or raster?
2. At what scale, and in which units?
3. Which files does it export, such as DWG, DXF, PDF, IFC or GLB?
4. Does it build on my venue file or redraw it?
5. What was the model trained on?
6. Are constraints enforced, or only written into the prompt?
7. Does a change re-run the checks?
8. Is the twin synchronized with the building, and how often?
9. Where does inference run, and is my data used for training?
10. Which actions does the agent take without my approval?

Hold the vendor to the word

Clear words protect designers from vague claims. Ask which of these forty terms describes what the tool really does. Then ask for the file.

Frequently asked questions

What is the difference between a raster image and a vector drawing?

A raster image is a grid of pixels. It carries no units, so you can only measure it roughly. A vector drawing stores lines as coordinates. A DWG or vector PDF in known units can be measured, scaled and edited. Any plan you build from should be vector.

Is a 3D virtual tour the same as a digital twin?

No. A tour shows what a place looked like on the day it was captured. A digital twin is kept in step with the real place at a stated rate and accuracy. That is why you can trust a twin for planning, not only for viewing.

What does hallucination mean in AI design tools?

It is output that looks right but is false. Examples are an invented door, a wrong ceiling height or a misread scale bar. Models state these as calmly as correct answers. So check every dimension that matters against the source drawing.

What is a diffusion model in simple terms?

It is an image generator that turns random noise into a picture, one small step at a time. A prompt guides each step. Most AI rendering tools use one. It makes convincing images of spaces, but it gives no measurements or geometry you can build from.

Monet works on the vector side of this glossary. It builds AI layouts to scale from your venue plan and brief, uploaded as a PDF or DWG. It exports them to DWG, DXF, PDF, JPG and GLB, and its venue digital twins show your own layout in the real room.