Ten Years of Design Experience Compressed into PPT Skills: A Major Update

After open-sourcing guizang-ppt-skill (github.com/op7418/guizang-ppt-skill) last time, everyone loved it. In just a few weeks, GitHub stars surged to over 6,000. It was also added to the highly popular open-source Claude Design references.

I have seen PPTs generated by this skill in numerous online and offline settings.

After the release, the most frequent issues received from users were:

• “Can you add more visual styles?”
• “Can you also handle the accompanying images?”
• “Do I need to redraw the cover after generating the PPT?”

I kept track of these issues while using the tool myself, accumulating feedback for two weeks. This time, I’m patching all these holes at once.

What’s New

A brand-new visual style has been added, image generation capabilities are directly integrated with Codex, and covers for platforms beyond PPT can now be created seamlessly.

Three specific updates:

• New Style B: Swiss Internationalism. Fully sans-serif, a single high-saturation anchor color, and grid-supremacy.
• Codex integration with GPT-Image 2.0. Directly generates images matching the tone: film-grain portraits, flowcharts, and beautified UI screenshots.
• Multi-platform cover generation. The same content can be assembled into covers for Xiaohongshu (Little Red Book), WeChat Official Accounts, Video Accounts, and more.

Triggering the New Style

Once installed, simply tell Claude or Codex: “Help me create a Swiss-style PPT.”

If you already have it installed, you can tell your AI: “Update guizang-ppt-skill for me.”

It will then ask you to choose from four theme palettes: Klein Blue, Lemon Yellow, Lemon Green, or Safety Orange.

• Klein Blue (IKB): Universal, business releases, AI products (default recommendation).
• Lemon Yellow: Youthful, sports, retail, Y2K retro.
• Lemon Green: Eco-friendly, sustainable, Gen Z branding.
• Safety Orange: Alerts, news, energetic themes.

As before, custom hex codes are not accepted.

I specifically wrote this rule into the hard constraints of SKILL.md. I explained the reason in the previous article, so I won’t repeat it here.

Key Layouts

Swiss style comes with 22 ready-to-use named layouts, covering covers, sections, data, comparisons, and conclusions.

Let’s highlight the six most distinctive ones:

• Cover: The left half features an IKB background with a line of large white text; the right half is left blank for meta information. Almost always used as the opening slide.
• Statement: A single sentence occupies 9.6vw, leaving only a small footnote. Ideal for starting sections or presenting core arguments.
• KPI Tower: Four solid-color bars with heights determined by data, with category labels below. Vertical numerical comparisons like profit margin tiers, pricing levels, or conversion funnels are instantly readable.
• Loop Diagram: Numbered steps distributed on concentric circles, forming a closed loop. Perfect for explaining self-learning loops, Agent automation cycles, or product feedback flywheels.
• Duo Compare: A thin line down the center, with text and data on each side. Use this for Old System vs. New Solution, Traditional vs. AI, or Before vs. After.
• Closing Manifesto: The left half features an IKB background with a white statement; the right half summarizes the deck with three takeaways. Recommended for the final slide to create a color loop with the cover.

Other layouts include horizontal timelines, loop diagrams, Three Forces confrontation, system hierarchy diagrams, “Why Now” three-point arguments, tech specs, and image heroes with KPIs.

Each layout corresponds to a typical content format. When you say “I need an industry ranking” or “I need a product benchmark,” the AI will automatically select the most suitable layout from the 22 options—you don’t need to memorize names.

After selecting a theme, the process remains the same as the previous version: six clarifying questions, followed by an outline and thematic rhythm table. Once aligned, the code is generated.

Let GPT-Image 2.0 Handle Your Images

If you are using the Codex environment, after the PPT is generated, it will proactively ask: “Do you want to generate some accompanying images for this PPT?”

If needed, it will automatically generate suitable images based on the selected PPT style and content:

• Humanistic documentary photos (film-grain texture)
• Infographics (flows, comparisons, system relationships)
• Redesigned screenshots (remaking your original images to fit PPT aspect ratios)
• Data posters, flowcharts, and system relationship diagrams

Generated images will automatically adapt to the current deck’s style and theme colors.

Here’s how it works:

• Electronic Magazine Style: Infographics use an e-ink tone—mostly black, white, and gray, with small amounts of low-saturation accent colors, thin lines, grids, and restrained whitespace.
• Swiss Style: Infographics follow Swiss modernism—sans-serif short labels in Helvetica/Inter style, 12/16-column grids, and right angles.
• Color Adaptation: If Klein Blue (IKB) is selected, the generated images use IKB blue as the sole anchor color. The same applies to Lemon Yellow, Lemon Green, and Safety Orange.
• Language: Text language follows the deck—Chinese decks use Chinese labels, English decks use English.
• Clean Output: Generated images do not include PPT shells (no headers, footers, page numbers, signatures, or decorative borders).

This detail ensures visual consistency across the entire deck without you having to monitor it.

You won’t encounter disjointed visuals where the PPT is IKB blue but the accompanying image suddenly turns green. You also don’t need to manually tweak prompts for GPT-Image while writing the PPT.

Visual drift across tools has been the most frustrating hidden cost in my two years of using AI for content creation.

By resolving this at the Skill level, users can worry less about one major pain point.

Generate Platform Covers

Before publishing a PPT, you always need to solve for three aspect ratios: WeChat Official Account header (21:9), Xiaohongshu portrait (3:4), and Video Account landscape covers.

Simply say: “Based on the core points of this PPT, give me a 3:4 Xiaohongshu cover,” or “Give me a 21:9 WeChat Official Account header.” The AI will generate images according to the Skill’s visual rules (same theme colors, same fonts, single-focus principle).

Batch generation is also supported. Under Xiaohongshu’s carousel mechanism, you can ask it to “batch generate 6 images with unified style, consistent font sizes, and varied layouts” to get them all at once.

WeChat Official Accounts are slightly different: the header is 21:9, but the share card uses a 1:1 square image. Both need visual continuity.

The approach is to generate them separately but using the same color palette and copy:

• First image: 21:9 main header, with the main title on the left and a visual anchor on the right.
• Second image: 1:1 square for the share card, extending the visual element of the header.

Two layout templates I commonly use: solid background + large text + date corner mark, or a split layout with images on the left and text on the right.

A Few Worth-Sharing Design Hacks

Hack 1: Use Film Grain to Combat the “AI Look”

Creating a presentation about “one person building a product” requires an image that conveys the emotion of “working alone.”

Previously, I’d spend half an hour flipping through Unsplash or use an obvious stock photo.

Now, I simply say: “Generate a 16:10 documentary photo with the theme of a person working alone in a studio late at night, natural light, low saturation, slight film grain, Fujifilm texture.”

The resulting image has a restrained human warmth and lacks that plastic “AI feel.”

I documented this judgment in references/image-prompts.md:

Film grain is the greatest value of GPT-Image 2.0 for PPT accompanying images. It removes the “AI-generated” label from the image.

Hack 2: Redesign Screenshots with Odd Aspect Ratios

Many content creators have a collection of original screenshots: product UIs, backend pages, data dashboards, all with varying aspect ratios and inconsistent whitespace.

Slapping them into a PPT immediately disrupts the visual flow. The old solution was Photoshop.

Now, you just feed the original image to Codex:

“Redesign this version to a 16:10 ratio, preserving all UI elements, adding a realistic work environment context, with medium visual density.”

It will regenerate an image that complies with PPT standards, retaining all key information with unified proportions and whitespace.

This is especially useful for product review content. Running all screenshots through GPT-Image 2.0 makes the entire PPT look like it was designed by one person.

Hack 3: “Wrap” AI Images in PPT Templates

This is a usage trick I discovered through testing.

If you post a GPT-Image 2.0 generated image alone, AI detection tools will likely flag it as “suspected AI-generated.”

On social media platforms, this can negatively impact recommendation weights.

However, if you place this image inside a PPT template and take a screenshot of the whole slide, the result is completely different.

Essentially, this assembles AI-generated assets into a complete work, adding a human touch.

It remains fact-based and non-fictional, simply avoiding being mistakenly penalized by algorithms.

Why It Looks Good

After covering the tactics, let’s discuss the design philosophy. Where do these Swiss-style visual rules come from?

My visual anchor is Massimo Vignelli plus Helvetica Forever.

It draws from the Swiss Internationalist tradition of the New York subway system, Unimark, and Müller-Brockmann.

If you’ve opened the book Helvetica Forever or seen Vignelli’s 1970 navigational system for the New York subway, you’ll recognize this language.

Comparing the two styles highlights the difference immediately:

Simply put: Style A is for storytelling; Style B is for presenting facts.

7 Design Disciplines

The visual rules of Swiss style actually boil down to seven principles, each rooted in industry tradition for a century.

My job was simply to write them into the Skill line by line, letting AI execute them for me.

1. Single Anchor Color: Only one highlight color is allowed per deck. If it’s blue, it’s only blue; if it’s yellow, it’s only yellow. No blue-yellow collage. This is the soul of Swiss style. “Less is more” isn’t just a slogan; it’s a hard rule written into the code.
2. Extreme Font Size Contrast: The ratio between main title and body text must be at least 8:1. Cover declarations use min(11.6vw, 19vh), while body text is 1.1vw. Visual tension comes from this contrast, not from decorations.
3. Larger Titles Should Be Thinner: Main title font weight should be 200 (ExtraLight). Do not use 700, 800, or 900. Large Swiss-style text is like a ruler on an architectural blueprint—it must be seen but not shouted. I learned this the hard way; initially using weight 800 made the slide look like a generic PowerPoint.
4. Right-Angle Solids: Cut out border-radius, box-shadow, and linear-gradient. All color blocks are right-angled, and all borders are 1px hairlines. It sounds harsh, but all that “consumer app feel” and “SaaS template feel” comes from abusing these three properties.
5. Grid Supremacy: A 16-column grid with 16px gap. All elements snap to the grid. Use left alignment and large whitespace for asymmetrical aesthetics. No centering, no even distribution.
6. No WebGL Backgrounds: Style A used WebGL fluids for a breathing effect; Swiss style intentionally removes this. Pure white is its background; any dynamic background is a distraction.

In Conclusion

The core of this update answers just one question: How long is the chain for human-AI collaborative content creation?

In the previous open-source release, I completed the “make a PPT” step.

This time, I connected “image generation” before it and “multi-platform covers” after it. Adding a new style and a new color loop closes the entire chain.

From writing outlines, generating PPTs, adding images, exporting, to publishing on different platforms—tasks that previously required opening five different software applications can now be completed in a single conversation.

Given the abundance of PPT Skills on the market, why does Master Zang’s PPT Skill still achieve such high usage and attention?

AI can only ever do 70% of the work. Every page layout in these two templates was fine-tuned manually by me on top of the AI foundation. In other words, I am designing using natural language.

Even in the AI era, 90% quality content remains precious.

The Skill has been updated on GitHub: github.com/op7418/guizang-ppt-skill

The update process is the same as last time. The README contains an “installation prompt for AI.” Copy and paste it to your Claude Code, Codex, or any AI Agent with shell access, and it will automatically pull the latest version.

Once installed, just say “Help me make a Swiss-style PPT” to trigger the new style.

If you found this helpful, please like and share. Feel free to post your PPTs or covers made with this Skill in the comments.