Concept Design and Massing Studies with AI

The concept phase is when AI earns its keep. You can spin up twenty site-response options in the time it used to take for five, explore programmatic alternatives while the client is still talking, and stress-test a massing scheme against solar orientation, setbacks, and precedent projects before you ever open Revit. The trick is learning which AI tools do which parts of the job — because generating pretty renders is the easy part, and generating defensible zoning analysis is the part that actually wins the commission.

What You'll Learn

• How to use AI for rapid programming and spatial diagrams
• The difference between image AI (Midjourney, DALL-E) and technical AI (ChatGPT, Claude) in concept design
• How to generate massing options that respect zoning, code, and context
• A workflow for moving AI concept output into Revit or Rhino without re-drawing everything

The Concept Phase AI Toolkit

Concept design uses three different types of AI, and mixing them up wastes time.

Technical chat AI (ChatGPT, Claude, Gemini): Best for programming analysis, zoning envelope calculations, precedent research, and textual description of a design concept. Treat these as your strategic design partner.

Image AI (Midjourney, DALL-E 3 in ChatGPT, Stable Diffusion, Adobe Firefly): Best for mood boards, facade exploration, and presentation renders. Treat these as your visual brainstorm partner — never as a final drawing.

Generative design plugins (Veras for SketchUp, Galapagos/Wallacei in Grasshopper, Forma from Autodesk, TestFit for feasibility): Best for parametric massing, zoning envelope optimization, and feasibility studies. Treat these as your parametric massing engine.

Most architects only use one of the three and miss 80% of the speed gains. The workflow below combines all three.

Step 1: Programming with Technical AI

Before you draw anything, use ChatGPT or Claude to pressure-test the brief.

Sample prompt:

Act as a senior architect. I have a client who wants a 45,000 sf suburban veterinary clinic on a 2.1-acre site in a B-1 zone in {city/state}. The program is: 12 exam rooms, 2 surgery suites, boarding for 40 animals, retail reception, staff break area, and outdoor play yards. Estimate the net-to-gross factor, suggest a reasonable adjacency diagram, and list the three programmatic conflicts I am most likely to hit (e.g., boarding noise vs. exam rooms). Flag anything that would typically need a special use permit.

Output: a programming table you can paste into a feasibility deck in five minutes. You still verify the NSF/GSF ratio and the zoning district.

Step 2: Zoning Envelope Analysis

For any site, ask AI to help you translate zoning code into a buildable envelope.

Summarize the dimensional standards of {zoning district} in {jurisdiction}: maximum height, front/side/rear setbacks, FAR, lot coverage, parking minimum, required landscaped open space. Then calculate the maximum buildable envelope for a {lot area} sf lot assuming {existing conditions}. Show the math and flag anything that requires a variance, special use permit, or zoning board approval.

This is a place where hallucinations happen frequently. Always confirm the zoning text by reading the actual municipal code — the AI is a faster first read, not a final authority.

Step 3: Massing Options with Generative Design

For feasibility and early massing, the best tools are parametric:

• Autodesk Forma (formerly Spacemaker): Generates and evaluates massing options against daylight, wind, noise, and view metrics. Strong for urban sites.
• TestFit: Generates feasibility layouts for multifamily, industrial, and parking — extremely fast for yield studies.
• Veras for SketchUp/Revit: Applies AI image style transfer to your massing to produce atmospheric renders.

Use these to produce options, then bring the chosen option into your main modeling tool.

Step 4: Facade and Material Exploration with Image AI

Once you have a massing, use Midjourney, DALL-E, or Adobe Firefly to explore expression.

Example Midjourney prompt:

a three-story mixed-use building on a tight urban corner, ground floor retail with tall glazing, second and third floors with operable windows in a pattern of vertical ribbons, warm brick and anodized aluminum, afternoon light, photographed by Iwan Baan --ar 3:2 --style raw

Tips for AEC image prompts:

• Name the project type, number of stories, and site context
• Describe materials in familiar architectural language (anodized aluminum, board-formed concrete, stucco, brick)
• Name a lighting condition (morning light, overcast, golden hour)
• Reference a photography style you admire (Iwan Baan, Hufton+Crow) to get a plausible composition
• Use --ar to set aspect ratio for plan-view, elevation, or perspective

Critical warning: Image AI outputs are not drawings. Do not extract dimensions. Do not trust windows, structural grids, or building systems shown. Use them as mood board images only.

Step 5: Moving from AI Concept to BIM

Architects constantly hit the "now what" problem: you have a great AI-generated concept image and a massing study, but you need to get it into Revit or Rhino for actual design development.

A practical workflow:

1. Pin the key moves in writing. Ask ChatGPT or Claude: "Summarize the essential design moves of this concept in 6 bullets (massing, circulation, primary materials, facade logic, roof, site response)." Print this and keep it next to your screen as you model.
2. Model only the primary gesture. In Revit, build the massing with in-place masses or generic walls. Do not model what you cannot defend.
3. Use AI as a co-modeler. Paste your massing plan into ChatGPT and ask "does this respect the setbacks I described earlier?" Use it as a sanity check, not a designer.

When AI Concept Work Has Limited Value

Be realistic about where AI stops helping:

• Complex client politics: If the project has three warring stakeholders, AI cannot navigate the meeting.
• Culturally specific or sacred projects: AI precedent is heavily Western-biased and shallow on non-Western typologies.
• Regulatory nuance: Waterfront, historic district, flood, wildfire, and airspace overlays almost always need human interpretation.
• True innovation: AI is a brilliant mimic, not an inventor. It will suggest what has been done; it will not suggest what has never been done.

A Concept Phase Checklist

Before you present any AI-assisted concept:

• Zoning numbers confirmed against the actual municipal code
• Program adjacencies reviewed against client program document
• Any AI-generated image clearly labeled as concept imagery, not a drawing
• Massing validated against solar, setbacks, and code
• Precedent references verified by opening the actual project page

Key Takeaways

• Use technical AI (ChatGPT, Claude) for programming and zoning analysis
• Use image AI (Midjourney, DALL-E) for mood boards only — never as drawings
• Use generative design tools (Forma, TestFit, Veras) for parametric massing
• Always verify zoning, code, and precedent claims against source documents
• Pin AI concept decisions in writing before modeling so the design logic survives the transition to BIM