@yaohui12138: Recently, big shots like guizang, zarazhangrui, and Hua Shu have open-sourced PPT Skills. I spent a day testing 7 PPT skill projects and here's my conclusion: The core reason why AI-generated PPTs look ugly is not that AI lacks ability, but that you haven't given it the right aesthetic constraints...

Recently, big shots like guizang, zarazhangrui, and Hua Shu have open-sourced PPT Skills. I spent a day testing 7 PPT skill projects and here's my conclusion: The core reason why AI-generated PPTs look ugly is not that AI lacks ability, but that you haven't given it the right aesthetic constraint system. These 7 open-source Skills solve exactly this—they compile "what kind of PPT humans are willing to look at" into rules that AI can understand. Below is a categorized recommendation by use case. Just pick the right one for you: 1. frontend-slides: A Vibe Coding example that lets you choose a style by looking at pictures GitHub: zarazhangrui/frontend-slides (17k+ stars) This is the one I tested that best aligns with the "Vibe Coding" philosophy. Key highlight: Instead of asking you what style you want, it directly generates 3 preview images in different directions for you to choose from. After you pick, it generates the full deck, with zero CSS/JS knowledge required throughout. Supports 12 carefully selected style templates, PPT to web, one-click Vercel deployment, or PDF export. Author zarazhangrui said bluntly on X: "Code-generated slides can be better than most PPT tools, but only if you first make Claude understand what 'beautiful' means." The essence of this Skill is Show, Don't Tell—instead of letting users describe aesthetics, it lets them point out aesthetics. Suitable for: situations where you need a visually designed presentation quickly, especially for external pitches or online sharing. 2. huashu-design: Generate an HTML deck + editable PPTX with a single sentence GitHub: alchaincyf/huashu-design (13k+ stars) Hua Shu's project is more ambitious—it's not just about making PPTs, but building an "HTML-native design system." A single prompt can output: - Professional HTML deck (opens directly in browser) - Editable PPTX (for colleagues who don't know code) - MP4 export (can be used as a video) - Interactive prototype (clickable App demo) Includes 20 design philosophies (Bauhaus, Swiss, Brutalism...), a 5-dimensional design review system. In the Chinese X community, it's often recommended alongside the guizang project. The two have different aesthetic routes: huashu leans toward the "disappearance of design tools" feeling, while guizang leans toward "magazine editorial aesthetics." Suitable for: scenarios requiring multi-format delivery, or collaboration between technical and non-technical team members. 3. guizang-ppt-skill: Specialized in curing AI-generated PPT aesthetic disasters GitHub: op7418/guizang-ppt-skill (7k+ stars) After sharing his PPT templates on X, Guizang directly open-sourced his entire aesthetic system as a Skill. This project has a clear positioning: horizontal page-turning magazine-style HTML PPT. - 10 layout skeletons, 5 theme color schemes (no custom hex values allowed, mandatory aesthetic protection) - WebGL fluid backgrounds, Motion One-driven entrance animations, single file output (works offline) Guizang wrote in the README: "If the color combination is wrong, the scene instantly turns ugly. Protecting aesthetics is more important than giving freedom." This hits the core pain point of AI-generated content—freedom and quality are often inversely related. Suitable for: 15-30 minute offline presentations or private sharing sessions where you need to highlight personal style. 4. open-slide: The Agent's last-mile productivity tool GitHub: 1weiho/open-slide (3k+ stars) This project has the most unique approach: not a "PPT generator," but a "Slide framework designed for Agents." Each slide is a React component with a fixed 1920x1080 canvas. Core workflow: - Generate the entire deck with a single prompt - Click any element in the browser to leave a comment - Run /apply-comments, and the Agent automatically applies the changes Supports speaker notes, timer, exports to HTML/PDF. Suitable for: scenarios requiring high-frequency iteration and multiple rounds of revision, or developers already using Claude Code/Cursor. 5. html-ppt-skill: The template enthusiast's arsenal GitHub: lewislulu/html-ppt-skill (3k+ stars) This project follows the "template richness" route: - 36 themes - 15 complete deck templates - 47 animation presets - Speaker mode (script + timer) Suitable for scenarios where you don't want to design from scratch and just use templates and modify content. If your need is "quickly produce a decent version," this library has the highest template density. 6. beautiful-html-templates: Agent automatically picks and fills GitHub: zarazhangrui/beautiful-html-templates (1k+ stars) Also by zarazhangrui, this is more of a template repository for frontend-slides. - 32 HTML Slide templates, prepared specifically for coding agents - The Agent can automatically select the appropriate template based on content type. Suitable for: batch generating multiple decks with consistent style, or establishing unified visual standards for a team. 7. open-design: Open-source alternative to Claude Design GitHub: nexu-io/open-design Tom Huang's team open-sourced this design system. It supports: - Export slides, images, videos, and HyperFrames - 71 brand design systems - Companion tool html-anything (converts multiple formats to high-aesthetic HTML, including Keynote-style PPT) This project's ambition is to be a "local-first alternative to Claude Design." If you need not just PPT but a complete design workflow, this is the most comprehensive. What these Skills have in common is that they are redefining "what a good-looking PPT is" with code. In the past, making a PPT meant dragging and dropping in PowerPoint; now, it's about describing what you want to the AI. But AI doesn't understand aesthetics, so what these open-source authors are doing is essentially compiling aesthetics into rules that AI can understand. That's the real value of these projects. If you're also using Claude Code for content production, these 7 libraries are worth cloning and trying out. Because you'll find: the ceiling of AI-generated content is often not in AI itself, but in how rigorous the "aesthetic constraint system" you give it is. All projects are open-source and free. Just search for the project name on GitHub. After installation, enter the corresponding skill name in your tool to use it.