AI-generated slides that actually look designed
A Python pipeline that turns documents into styled presentation images and optional AI-generated transition videos.

What it does
NanoBanana PPT Skills reads a document or JSON plan, extracts structure with an LLM, and generates 16:9 slide images via Google’s Gemini image model. It can also stitch in AI-generated transition videos using Kling AI, then export the whole sequence as a single video via FFmpeg. Two built-in visual styles are offered: a “gradient glass” tech look and a flat vector illustration style.
The interesting bit
The project is aggressively optimized for Claude Code: it ships with copy-paste prompts that ask Claude to install itself as a Skill, manage API keys, and even generate the transition prompts between slides. It’s less a standalone app and more a conversation starter for your agent.
Key highlights
- Generates 2K or 4K slide images from a JSON plan or natural language input
- Optional transition videos via Kling AI API, with resolution/framerate normalization
- Interactive HTML player for keyboard-navigated preview
- Packaged as a Claude Code Skill with
/ppt-generator-procommand - Requires Google AI API key (images); Kling keys are optional (video)
Caveats
- README is entirely in Chinese; English speakers will need translation
- Video generation requires a second paid API (Kling) and FFmpeg installation
- “2K ~30 seconds per page” claim is stated without methodology or reproducible benchmark
Verdict
Worth a look if you live in Claude Code and need quick, visually consistent slide decks without opening Figma. Skip it if you need fine-grained layout control or can’t stomach two separate AI API bills.