Banana Slides – AI‑native PPT creator

What it is

• A web/desktop application that turns ideas, outlines, documents or images into fully‑editable PowerPoint (PPTX) presentations.
• It uses large‑language‑model (LLM) providers (Gemini, OpenAI, Claude, etc.) and image‑generation models to produce slide content and graphics.

Key capabilities

Feature	What you can do
Prompt‑to‑slide generation	Type a one‑sentence idea or upload a document; the AI creates a slide‑by‑slide outline and fills each slide with text and images.
“Vibe” natural‑language editing	Speak or type commands like “replace the chart on slide 3 with a pie chart”; the model updates the slide in real time.
Template control	Upload your own PPT/PDF template or reference images; the system parses the style and automatically matches generated slides to it, or you can bind templates per‑slide.
Material parsing	PDFs, DOCX, Markdown, plain‑text files are parsed; text, images and tables are extracted and added to a project‑wide asset library.
Export options	One‑click export to editable PPTX (beta) or PDF; optional slide‑transition animations; also can render a narrated video (MP4) with AI‑generated voice‑over and subtitles.
Project history & reproducibility	Every edit is stored; you can revisit a preview link or, in the self‑hosted version, view a history page protected by a password.
Deployability	Provided as Docker images (frontend + backend) and as a desktop build (PyInstaller). A one‑click “rain‑cloud” deployment template is also offered.

How it works (high‑level)

1. Front‑end (React/Vite) collects user prompts, uploaded files, and template choices.
2. Back‑end (Python, uv‑managed) routes requests to the selected LLM provider (Gemini, OpenAI, Claude, etc.) for text generation and to an image‑generation provider (e.g., Sensenova U1, APIMart) for graphics.
3. The service parses uploaded documents, stores extracted assets, and keeps a SQLite database of project metadata.
4. Generated slides are assembled into a PPTX using python-pptx; for the editable export the images are rendered with a clean background so text can be edited later.
5. Optional TTS (text‑to‑speech) creates a voice narration; FFmpeg combines it with slide images and subtitles into an MP4 video.

Getting started

• Quick demo: https://bananaslides.online/ (uses fixed API keys for a few models).
• Self‑host: clone the repo, copy .env.example → .env, fill in API keys for the LLM/image services you want, then run docker compose up -d (or build locally). Frontend runs on port 3011, backend on 5011.
• Desktop version: download the RC7 release (v0.9.0‑rc.7) and run the packaged executable; it checks for updates automatically.

Who it’s for

• Non‑designers who need a polished deck quickly.
• Professionals looking for AI‑generated layout ideas to speed up slide creation.
• Educators & students who want to turn lecture notes or assignments into visual presentations.
• Developers who can extend the platform (open API, Docker, source‑available).

Roadmap highlights

• Multi‑layer precise cut‑out for fully editable PPTX (in progress).
• Integrated web search for up‑to‑date facts.
• Agent‑mode workflows that let the AI orchestrate multiple steps (e.g., fetch data → generate slides → produce video).

Licensing & community

• Open‑source under the MIT license.
• Active GitHub repo with stars, forks, and a public Discord/WeChat group for support.
• Sponsorship options listed (AIHubMix, APIMart, Volcengine) for those who want to contribute financially.

---

All details are taken directly from the repository’s README; no additional features are assumed.