Lab & ConceptsLab & Concepts
存放尚未长成完整产品的方向实验，用更低成本测试交互方式、信息结构和可行性。A place for ideas that are not ready to become full products, but still deserve a small test.
- 工作流Workflow
- 原型Prototype
- 参数化设计Parametric design
- AI SkillsAI Skills
背景与初衷Context
很多想法在最初阶段并不知道是否值得产品化。先做一条工作流、一个可运行原型或一套规则，比一开始就搭完整系统更容易看见真实问题。At the beginning, many ideas do not yet justify a full product. A workflow, runnable prototype, or constrained ruleset reveals the real problem sooner than a complete system.
AI Skill 实验AI skill experiments
节点式生成Node-based generation
ComfyUI 工作流实验ComfyUI workflow studies
把生成、修复、放大与自动化步骤整理成可见节点链路。Organizes generation, restoration, upscaling, and automation as a visible node graph.
多模态生成Multimodal generation
建筑视觉提示词实验Architectural prompt studies
对照提示词结构、参考图和候选结果，验证视觉方向的可控性。Compares prompt structures, references, and results to test visual controllability.
GrasshopperGrasshopper
参数化古建生成Parametric heritage generator
把四合院与传统木构规则写入参数定义，用数据驱动三维模型。Encodes courtyard and timber construction rules in a data-driven 3D definition.
Coze 工作流Coze workflow
个人 AI 意图路由Personal AI intent router
识别来访者意图，分流到产品知识、个人项目、联系方式或自我介绍。Routes visitor intent to product knowledge, projects, contact details, or an introduction.
术数工程化Auditable domain rules
Numerologist SkillsNumerologist Skills
用固定脚本、规则集与参考资料约束奇门遁甲和紫微斗数的计算与解释。Uses deterministic scripts, rulesets, and references to constrain calculation and interpretation.
查看仓库View repository
视觉提示词系统Visual prompt system
Phil Design SkillsPhil Design Skills
19 套为 gpt-image-2 编写的视觉风格 Skills，可把文章、主题、产品或已有资产编译成完整提示词。Nineteen visual-style skills for gpt-image-2 that compile articles, themes, products, or assets into complete prompts.
查看仓库View repository核心痛点Problems
新工具和交互方向需要先用小实验验证，再决定是否投入完整产品。New tools and interaction ideas need a small test before they deserve full product investment.
- AI 生成结果容易受提示词、模型和工作流节点变化影响。AI outputs can change sharply with prompts, models, and workflow nodes.
- 专业领域规则若只写在自然语言里，很难稳定复现和检查。Domain rules written only in natural language are difficult to reproduce and audit.
- 过早包装成完整产品会放大实现与维护成本。Premature product packaging increases implementation and maintenance cost.
处理方式Approach
针对一个具体问题建立最小实验，记录输入、约束、处理步骤和输出，再根据可控性与复用价值决定是否继续。Build the smallest experiment around one problem, record its input, constraints, steps, and output, then continue only when control or reuse value is clear.
目标Goal
用小规模、可回看的实验判断一个方向是否值得继续。Use small, reviewable experiments to decide whether a direction deserves further work.
产品设计Product design
- 让复杂流程以节点、参数或脚本形式可见。Make complex processes visible as nodes, parameters, or scripts.
- 把领域知识拆成规则、参考资料与可执行步骤。Separate domain knowledge into rules, references, and executable steps.
- 保留实验截图和公开仓库，让结果可以回看与复用。Keep screenshots and public repositories so results can be reviewed and reused.
技术与实现Build notes
实验按问题选用 ComfyUI、Grasshopper、Coze、脚本与 Agent Skills；共同方法是先固定输入和边界，再检查中间步骤与最终输出。Experiments use ComfyUI, Grasshopper, Coze, scripts, and agent skills according to the problem. The shared method is to fix inputs and boundaries first, then inspect intermediate steps and final output.