界面已本地化;技能介绍与说明文档按所选语言由 AI 自动翻译,代码、命令、路径和专业术语保留原文,以原作者文档为准。
探索技能
设计与创意 · SKILL.md

generative-ai-design

generative-ai-design

【设计与创意】Generative design for construction: text-to-BIM concepts, option generation, and AI-assisted design iteration with cost and carbon feedback. Use when exploring early design options with AI.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: "generative-ai-design" description: "Generative design for construction: text-to-BIM concepts, option generation, and AI-assisted design iteration with cost and carbon feedback. Use when exploring early design options with AI."

Generative AI Design for Construction (2026)

What is real in 2026

Generative design in construction is option generation with feedback, not autonomous architecture: given site constraints, program and budget, an AI generates massing/typology options and scores them on cost, carbon and buildability — the human designer selects and refines.

The loop

Constraints (site, program, budget)
        │
        ▼
Generate options (LLM/parametric/optimisation)
        │
        ▼
Quantify each option (BIM takeoff + CWICR cost + carbon)
        │
        ▼
Score & rank (cost/m², kgCO₂e/m², GFA efficiency)
        │
        ▼
Human selects → refine → detail

Toolchain

StageTools
Massing generationparametric tools (Rhino/Grasshopper, Dynamo) + LLM sketches
Text-to-conceptimage models (Midjourney/DALL·E) for moodboards; text-to-BIM is early-stage (Hypar, Finch, qbiq)
QuantificationOpenConstructionERP BIM takeoff (oce-bim-takeoff)
Cost scoringCWICR cost bases (oce-load-cost-bases)
Carbon scoringembodied-carbon-esg

Prompt pattern for concept generation

"Generate 3 massing options for a 12,000 m² residential building on a 30×60 m
plot, 6 storeys, max 40% glazing, Berlin climate. For each: GFA, FAR,
indicative structure, kgCO₂e/m² (A1-A3), €/m² construction cost."

Then quantify and rank:

OptionGFAFARCost/m²kgCO₂e/m²Verdict
A11,8002.91,050 €310lowest cost
B12,4003.11,180 €285lowest carbon
C12,1003.01,120 €295balanced

Guardrails

  • AI options are starting points, always human-reviewed and code-checked.
  • Cost/carbon scores come from real databases (CWICR + EPD), not LLM guesses.
  • Keep every option's inputs logged (reproducibility, AI Act transparency).
  • Text-to-BIM models are not yet permit-grade — treat outputs as concepts.

Resources

我的收藏