枋程设计BrianK
先用多轮追问把场地、风格和限制条件问清楚，再把生成步骤放到可检查的流程里。Clarifies site, style, and constraints before generation so concept work does not depend on one loose prompt.
- 任务书分析Brief analysis
- 概念收敛Concept convergence
- 多模型路由Multi-model routing
- A3 / PDF 导出A3 / PDF export
背景与初衷Context
建筑学生和设计师在概念阶段常同时面对模糊任务书、零散参考图与紧迫交付。通用生图工具能快速给图，却很难保留设计约束、解释方向选择并组织最终成果。Architecture students and designers often begin with an ambiguous brief, scattered references, and a tight deadline. Generic image tools produce images quickly but rarely preserve constraints, explain direction choices, or organize the final presentation.
打开项目Open project
可通过下方链接查看当前可访问版本。Open the current version through the link below.
核心痛点Problems
输入零散、方案难收敛，最终表达还要在多套工具之间反复搬运。Inputs are fragmented, concepts are hard to converge, and final presentation requires too much tool switching.
- 场地、功能与风格约束难以从混合输入中一次整理清楚。Site, program, and style constraints are difficult to extract from mixed inputs.
- 大量生成结果缺少收敛路径，下一轮修改没有依据。Large batches of generated images provide no clear path toward a decision.
- 分析、出图与排版彼此割裂，截止前仍要重复搬运图文。Analysis, generation, and layout remain disconnected, creating repeated work before delivery.
处理方式Approach
用四阶段路径组织工作：分析任务书并给出候选概念、深化选定方向、生成风格图与体块分析、汇入可编辑展板并导出 PDF。Use four stages: analyze the brief and propose concepts, develop the selected direction, generate style and massing visuals, then assemble an editable board and export a PDF.
目标Goal
验证一条从需求理解到可讨论、可继续修改、可导出成果的建筑概念最小闭环。Test a minimum viable workflow from requirement understanding to a concept that can be discussed, edited, and exported.
产品设计Product design
- 接收文字、图片与 PDF，并整理目标、场景、硬约束和禁用项。Accept text, images, and PDFs, then structure goals, context, hard constraints, and exclusions.
- 先展示候选概念和批判意见，确认方向后再进入生成。Show candidate concepts and critiques before generation so the user can select a direction.
- 将鸟瞰、室内、体块与分析图放进共享上下文的生成工作台。Keep aerial, interior, massing, and analysis generation inside one workspace with shared context.
- 提供可替换图文的 A3 竖版与横版展板，并导出 PDF。Provide editable A3 portrait and landscape boards with PDF export.
技术与实现Build notes
前端使用 Next.js 14、Tailwind CSS 与 Zustand；文本分析优先走 DeepSeek，图像理解与生成按场景使用豆包或 Gemini；Supabase 提供账号与数据能力，展板通过 html2canvas 和 jsPDF 导出。The product uses Next.js 14, Tailwind CSS, and Zustand. DeepSeek handles text-first analysis, Doubao or Gemini supports multimodal analysis and generation, Supabase provides account and data capabilities, and html2canvas plus jsPDF handles board export.