As AI tools have become routine in design and architecture practice, clients have begun using AI vocabulary in project conversations — sometimes precisely, sometimes incorrectly, often based on what they’ve read in a news article. Designers and architects who can translate between technical AI terms and client-friendly language win trust and avoid the awkward moments when a client asks about “text-to-image” or “training data” and gets either jargon or a blank stare. This article defines 25 AI design terms in client-friendly language — what each means, what it actually does in a real project, and the misconceptions worth correcting.
How do you use this glossary?
Each entry follows the same structure: short client-friendly definition, what it actually does in design work, and any common misconception worth catching. The terms are grouped loosely by concept area rather than alphabetically — the groupings help when a client asks about one term and the next question follows naturally.
Generation and Output
1. Text-to-Image
Client-friendly: Software that creates a picture from a written description.
In design work: You type “Mediterranean villa with terracotta roof and white plaster walls, late afternoon light,” and the software produces an image matching that description. Tools like Midjourney, Nuit (text-first mode), DALL-E, and Stable Diffusion all do this.
Common misconception: That text-to-image produces a buildable design. It produces an image, not a buildable specification. The architect still has to turn it into a real design.
2. Image-to-Image
Client-friendly: Software that takes an existing image and produces a modified version of it.
In design work: Upload a photo of your house, the tool restyles it. Upload a SketchUp viewport, the tool turns it into a finished-looking rendering. Tools like InteriorAI, REimagineHome, Veras, and Gendo work this way.
Common misconception: That image-to-image perfectly preserves the original. It mostly preserves composition but can change details — windows can shift, dimensions can drift, materials can swap unintentionally.
3. Sketch-to-Render
Client-friendly: Software that turns a hand drawing or rough sketch into a finished-looking image.
In design work: Architect sketches a building or interior, drops it into the tool, gets back an atmospheric rendering. Gendo is the best-known tool; mnml.ai, Veras, and others do versions of this.
Common misconception: That sketch-to-render is more accurate than text-to-image. It’s more compositionally controlled but still adds interpretation to whatever the sketch contained.
4. Generation
Client-friendly: A single output produced by an AI tool — typically one image, one plan, or one variant.
In design work: “I generated five concepts” means five separate outputs. Most pricing is per-generation or per-credit.
5. Variant
Client-friendly: A close alternative of a previous output — same direction, small change.
In design work: After generating a strong concept, designers often generate variants — different cladding, different window pattern, different time of day — to refine the direction without abandoning it.
6. Iteration
Client-friendly: The back-and-forth of generating, evaluating, refining, generating again.
In design work: Most design work with AI is iterative — first generation rarely lands; refinement through five to twenty iterations is normal.
Briefing and Control
7. Prompt
Client-friendly: The written description given to the AI to produce an output.
In design work: A well-written prompt typically includes style direction, materials, lighting, atmosphere, and a distinctive element. “Modern farmhouse” is a weak prompt; “two-story modern farmhouse with white board-and-batten siding, black standing-seam metal roof, large covered front porch with square columns, dark Dutch front door” is a strong prompt.
Common misconception: That short prompts produce good output. Specific prompts produce specific output; generic prompts produce generic output.
8. Prompt Engineering
Client-friendly: The skill of writing prompts that produce useful results.
In design work: Designers learn through practice what phrasings produce strong output for their typical work. There are no universal rules; it’s empirical skill.
9. Reference Image
Client-friendly: An image given to the AI alongside the prompt, used to guide style or composition.
In design work: Upload a photo of an existing building (to match style in an addition), a precedent image (to match atmosphere), or a previous generation (to maintain direction). Many tools accept reference imagery.
10. Style Preset
Client-friendly: A pre-built style direction the AI applies — typically named (“modern farmhouse,” “Japandi,” “Scandinavian,” “Mediterranean”).
In design work: Consumer-facing tools (InteriorAI, HomeDesigns.ai, ArchitectGPT) lean on style presets. Professional tools tend to use open prompts with named references rather than presets.
11. Negative Prompt
Client-friendly: A description of what the AI should not include.
In design work: “No people, no cars, no clutter, no signage” added to a prompt to keep the output clean. Some tools support this explicitly; others handle it through phrasing.
Coherence and Context
12. Project Context
Client-friendly: Information about your specific project carried across multiple AI generations to keep them coherent.
In design work: A tool with project context (Nuit) lets you pick an exterior, then generate interiors that match it, or generate a floor plan using an AI floor plan generator that stays coherent with the rest of the project. Tools without project context (Midjourney) generate each image independently, and coherence is your job.
Common misconception: That all AI tools maintain coherence across multiple generations. Most don’t — coherence is a specific tool capability.
13. Branching
Client-friendly: The ability to develop multiple directions in parallel from a starting point, like branches of a tree.
In design work: Generate six exterior concepts, pick two favorites, generate four refinements of each. Each path is a branch; you can pursue multiple branches without losing the others. Nuit’s interface is built around this; most other tools are linear.
14. Image Coherence
Client-friendly: The visual consistency of style, materials, and atmosphere across multiple images of the same project.
In design work: A client deck where every image reads as the same project — same light, same materials, same atmospheric direction. Coherence is harder than producing single beautiful images.
Models and Training
15. Foundation Model
Client-friendly: The large general-purpose AI that produces output. Most architectural AI tools are built on top of foundation models.
In design work: Stable Diffusion, FLUX, and various proprietary models are foundation models. Tools like Nuit, Midjourney, and Veras are products that combine foundation models with architectural-specific features.
16. Training Data
Client-friendly: The images and text the AI learned from — typically millions of images scraped from the public web.
In design work: AI tools learned what “Mediterranean villa” looks like by training on many images labeled with that term. Training data quality and scope affect output quality.
Common misconception: That AI “knows” specific architectural projects. AI has seen patterns across many images; it doesn’t have specific knowledge about your favorite Mies van der Rohe house unless trained explicitly on that data.
17. Fine-Tuned Model
Client-friendly: A foundation model that’s been additionally trained on a specific category — architectural imagery, interior design, technical drawings.
In design work: Specialized architectural AI tools often use fine-tuned models. They produce more on-target output for architectural work than general models like raw Midjourney.
18. Hallucination
Client-friendly: When AI confidently produces output that’s plausible but wrong — a chair with three legs, a staircase that doesn’t make structural sense, a faucet floating above a sink.
In design work: AI never refuses; it always renders something. Architects learn to spot hallucinations — small wrongnesses in otherwise good output. Hallucination is one reason AI imagery is concept-stage, not construction-stage.
Editing and Refinement
19. Inpainting
Client-friendly: Painting over a portion of an image and asking the AI to fill it in differently.
In design work: Highlight a cabinet in a kitchen rendering, ask for it to be painted blue instead of white. The rest of the image stays the same. Several tools support this.
20. Outpainting
Client-friendly: Extending an image beyond its original frame — adding more of the room, more of the building, more of the landscape.
In design work: Crop too tight on a beautiful rendering? Outpaint to extend the view. Less common than inpainting but useful.
21. Image Edit Tool
Client-friendly: A tool focused on making small targeted changes to existing AI images while preserving most of the image.
In design work: Nano Banana is the most respected tool in this category. Used heavily by professionals to swap a stone, change a fixture, recolor a wall, or move an element without regenerating the whole image.
Quality and Output
22. Aspect Ratio
Client-friendly: The proportion of the image — wide, tall, square.
In design work: Different uses want different ratios. Hero presentation imagery often wants wide; mobile-listing photos want square or vertical; magazine spreads want specific ratios. Most tools support multiple aspect ratios.
23. Resolution
Client-friendly: How large and detailed the output image is.
In design work: Web presentation imagery doesn’t need huge resolution; printed material does. Some tools generate at lower resolution and upscale; some generate natively at high resolution.
24. Upscaling
Client-friendly: Increasing the size and detail of an image after generation.
In design work: Many AI tools generate at moderate resolution; upscaling increases the image size for print or detailed presentation. Most tools include upscaling features.
25. Render Style
Client-friendly: The overall look of the output — photorealistic, illustrated, sketchy, painterly.
In design work: Photorealistic for presentation imagery; sketchy or painterly for concept exploration. Many AI tools support different render styles for different purposes.
Extended Glossary: Four Terms in Growing Use
26. AI Floor Plan Generator
Client-friendly: Software that creates a building layout diagram from a text description or dimensional inputs.
In design work: AI floor plan generators either take structured parametric inputs (lot size, room count, adjacencies) and produce dimensioned schematic plans — like Maket — or generate plans as part of a connected project workflow alongside exterior and interior views, like Nuit’s AI floor plan generator. Plans produced are concept-level: useful for program communication, not for construction.
Common misconception: That AI-generated plans are construction-ready. They show spatial intent and room relationships reliably; they don’t carry structural, code, or MEP coordination.
27. AI Interior Design Tool
Client-friendly: Software that generates or restyle interior visualizations from a text description or photo.
In design work: AI interior design tool products range from photo-restyling tools (upload an existing room, see it in a different style) to brief-driven generation tools that produce room concepts from text without a photo to start. Which type you need depends on whether you have an existing room or are designing from scratch.
Common misconception: That all interior AI tools work the same way. Photo-restyling tools require a room to start from; brief-driven tools generate from nothing. Using the wrong type wastes time.
28. AI Exterior Design
Client-friendly: Using AI software to generate or restyle the outside of a building — facade, massing, cladding, landscaping.
In design work: AI exterior design tools fall into two categories: generative tools that produce an exterior from a text brief (used in early concept work, before any sketch exists), and rendering tools that take a 3D model or sketch and apply a style to it (used in presentation phases). The distinction matters — using a rendering tool when no model exists, or a concept tool when a model already does, is a category mismatch.
Common misconception: That all exterior AI tools generate from text. Most require a model or sketch input; text-to-exterior tools are the minority.
29. AI Masterplan Generator
Client-friendly: Software that helps visualize the layout of a large site with multiple buildings, roads, open spaces, or zones.
In design work: AI masterplan generator tools help architects and developers produce early site-level layout concepts for multi-building developments, campuses, or mixed-use projects. Unlike single-building concept tools, they work at the site scale — how buildings relate to each other, how circulation works, where open space sits. Outputs are schematic; detailed masterplanning still requires specialist urban designers and engineers.
Common misconception: That single-building concept tools scale up to masterplan work. The spatial logic is different enough that specialized tools or separate workflows are needed at the site scale.
Translating Client Questions
A few common client-side questions and how to answer them in client-friendly language.
“Are you using AI?” “Yes — we use AI tools to explore concept directions faster and to produce atmospheric imagery. Our team makes the design decisions, refines the output, and develops the buildable design in traditional ways. AI compresses our concept phase; it doesn’t replace our design judgment.”
“Did the AI design my house?” “No. The AI helped us visualize directions quickly during the concept phase. The design — what works for your program, your site, your budget, your code requirements — is our work. The buildable drawings and specifications we produce are entirely our work.”
“Could I just use AI myself?” “You can certainly explore directions with consumer AI tools — and many of our clients do. What we add is design judgment, site knowledge, code expertise, technical coordination, and the production of buildable documents. AI doesn’t replace any of that.”
“How do I know the AI imagery is accurate to what we’ll build?” “The concept imagery shows direction — atmosphere, materials, character. The buildable details — dimensions, structural design, MEP, code compliance — are developed in design development and construction documentation. Some specific elements in the renderings will change as those details resolve.”
“Who owns the AI imagery?” “It’s part of our instruments of service, like all our other design work — covered by the same ownership and license terms in our contract.”
“Will AI replace architects?” “AI changes what architects do — concept exploration is faster, atmospheric rendering is cheaper — but doesn’t replace the architect’s role. Code knowledge, professional liability, multidisciplinary coordination, and construction administration all remain human professional work.”
Related reading
- AI Architecture Glossary: 50 Terms — The AI architecture space in 2026 has its own vocabulary — part machine learning, part…
- AI Client Presentations Without Slop — Clients in 2026 can recognize generic AI imagery in seconds — and the moment they think…
- What Is AI Concept Design? Definition — AI concept design is the use of generative AI tools to produce early-stage architectural…
- State of AI in Architecture: 2026 Annual Report — By mid-2026, AI tools are routine in architectural concept exploration, atmospheric…
- AI Architecture Design: The Complete Guide for 2026 — AI architecture design is the use of generative AI tools to produce architectural…
- AI for Architecture Students in 2026 — How students should approach AI tools: portfolio, studio projects, and thesis work…
Frequently Asked Questions
Why do I need to know AI vocabulary as an architect or designer?
Because clients are using it. Even clients without design backgrounds have read articles about AI tools and use the vocabulary in conversations. Being able to translate technical terms into client-friendly language builds trust; using jargon or freezing builds distance.
Which terms are most important to know in 2026?
Text-to-image, image-to-image, prompt, reference image, hallucination, project context, iteration, and image edit tool. These come up most often in client and team conversations.
Should I avoid technical terms entirely with clients?
No — clients respond well to technical literacy from designers. The skill is in calibrating: using technical terms when they help precision, translating to client-friendly language when terms might confuse. Don’t pretend the technical vocabulary doesn’t exist.
How is AI vocabulary evolving in 2026-27?
Continued addition of new terms as tools evolve. Look for vocabulary around 3D-model generation, real-time co-design, and AI integration with VR/AR walkthroughs to become more common through 2027.
What’s the difference between text-to-image and sketch-to-render?
Text-to-image starts from a written description; sketch-to-render starts from a visual input (sketch or model viewport). Both produce images. Sketch-to-render preserves composition more reliably; text-to-image allows more open exploration.
Why does the AI “hallucinate” wrong details?
AI generates by pattern-matching what it learned from training data. When asked for something complex or unusual, it interpolates — sometimes producing plausible-looking but wrong details (extra fingers, impossible structures, floating fixtures). Architects learn to spot and correct these in client deliverables.
What’s the most important thing for clients to understand about AI design?
That AI is a tool for concept exploration and atmospheric rendering, not a replacement for design or for buildable documentation. The architect or designer’s judgment, technical work, and professional responsibility remain central.
Try Nuit free — 100 credits, no card required. Generate coherent project concepts with the project context and branching workflow that distinguishes professional design work from generic AI output. Start your project →