How to Use NanoBananaPro to Generate AI-Powered PPT Slides

Design Overview

Generating a PowerPoint (PPT) with AI is split into two stages: (1) produce an outline and slide‑level textual content using a large language model, and (2) feed that content to an image‑generation model (NanoBananaPro) to create visual slides. Consistent styling is achieved by adding style cues to the image prompts.

Open‑Source Project: banana‑slides

The repository https://github.com/Anionex/banana-slides (≈12 K stars) implements a native AI PPT generator based on NanoBananaPro. It supports three creation paths— idea , outline , and page description —and can automatically extract assets from PDF, DOCX, Markdown, or plain‑text files.

Key capabilities:

One‑sentence generation : Provide a theme; the AI returns a structured outline with per‑page descriptions.

Natural‑language editing : Voice‑style commands (e.g., “change the third slide to a case study”) adjust the outline in real time.

Outline/description mode : Batch generation or manual fine‑tuning.

Rich asset parsing : Upload PDF/Docx/MD/Txt; the backend auto‑extracts key points, images, and charts.

Style reference : Upload reference images or templates to customize slide aesthetics.

Export formats : One‑click export to PPTX or PDF with a default 16:9 layout.

Editable PPTX (beta) : Export high‑fidelity, editable slides.

Local Deployment

Clone the repository and prepare the environment:

git clone https://github.com/Anionex/banana-slides
cd banana-slides
cp .env.example .env

Edit .env to set the required API keys and optional timeout/retry parameters, for example:

GOOGLE_API_KEY=YOUR_KEY
GOOGLE_API_BASE=https://api.apiyi.com
OPENAI_API_KEY=YOUR_KEY
OPENAI_API_BASE=https://api.apiyi.com/v1
OPENAI_TIMEOUT=300.0
OPENAI_MAX_RETRIES=2

Start the service with Docker (pre‑built images are available on Docker Hub): docker compose -f docker-compose.prod.yml up -d After the containers are running, open http://localhost:3000 to access the web UI, which mirrors the online demo.

Generating a PPT

In the UI, click “Next” to generate an outline, edit as needed, then generate page descriptions, and finally produce slide images. The resulting slides can be exported as PPTX or PDF with a single click.

Model Update: NanoBanana 2.0

Google released NanoBanana 2.0 (model name gemini-3.1-flash-image-preview). It offers Pro‑grade quality, Flash‑grade speed, 4K output, 14 aspect ratios (including 1:4, 4:1, 1:8, 8:1), and grounding image search. Pricing is $0.045 per request or $0.018 per image at volume. To switch models, change the IMAGE_MODEL variable in .env to gemini-3.1-flash-image-preview.

Custom Implementation on Baidu Miaoda

A similar AI PPT generator was built on the Baidu Miaoda platform, using Supabase for the backend and defining four edge functions for text generation, image generation, task querying, and result retrieval.

Prompt example for the text model (Chinese):

请帮我设计并规划一个功能完备、体验流畅的AI生成PPT的Web应用程序。该产品旨在通过先进的AI技术，将用户简单的文本描述快速转化为专业、美观的演示文稿。

The workflow mirrors the open‑source approach:

User inputs a theme.

Text model generates an outline (up to 8 slides) and detailed content.

User reviews and confirms the outline.

Image model generates visual assets for each slide.

The system packages the slides into a downloadable .pptx file.

Edge functions enable asynchronous handling of text and image generation tasks, allowing the front‑end to poll for status updates.

Conclusion

The banana‑slides project and the custom Miaoda implementation both demonstrate that NanoBanana models can serve as the core engine for end‑to‑end AI‑generated PPT creation, turning a textual idea into a full presentation with minimal manual effort.