Prompt engineering for AI slides that don't look like AI slides

What it does This repo collects and categorizes elaborate system prompts for NotebookLM (and the similar Kael.im service) that force the AI to generate slides in specific visual styles — from Swiss-style editorial minimalism to manga, streetwear, and neon vaporwave. Each prompt is a small design specification, dictating color palettes, typography ratios, grid systems, and layout rules.

The interesting bit The prompts read like briefs you’d give a human art director: “extreme jump ratio of 10:1 between headlines and body text,” “ban simplistic centered alignment,” “place the conclusion with a ‘bam!’ in the center.” It’s essentially prompt engineering as graphic design by proxy — users who can’t draw are writing enough constraints to constrain an LLM into taste.

Key highlights

• ~20 distinct style categories, from “Sharp-edged Minimalism” to “Deformed Flat Persona” and “Pink Street-style”
• Prompts sourced from Japanese and Chinese design Twitter/X, WeChat, and RED (Xiaohongshu) creators
• Each style includes example screenshots and the full prompt text, often with specific hex codes and font weights
• Companion project: a citation-checker skill that flags hallucinated or missing citations in generated slides
• CC BY 4.0 license — explicitly meant to be reused and remixed

Caveats

• The README is heavy on promotion for Kael.im (the author’s competing service) and light on provenance for some prompts
• Several prompts contain broken English and contradictory instructions that may confuse rather than guide
• No testing methodology or consistency metrics — “field-tested” is claimed but not demonstrated

Verdict Grab this if you’re already using NotebookLM and tired of its default beige corporate templates. Skip it if you need rigorous, benchmarked prompt engineering or if you find the “creative underground” framing a bit much.