Banana Slides

Banana Slides is an AI-native presentation generator maintained by Anionex. It turns a one-sentence idea, an outline, or page-by-page descriptions into full slide decks in which text, graphics, and layout are rendered together by image-generation models. It is available as a hosted demo, as a self-hosted Docker or source deployment, and as a desktop release candidate.

What It Is

Banana Slides is a slide-creation app built around image-generation models such as nano banana pro. Instead of filling fixed templates, each slide is rendered as a unified visual. Users can guide the look with reference images, uploaded templates, or a written style description.

Users can start from a topic, an outline, or detailed page descriptions. They can also upload PDF, DOCX, Markdown, or TXT files so that text and images are extracted and used as source material. Existing PPTX or PDF decks can be redesigned through a renovation flow.

Editing and Export Workflow

Changes are made in natural language: users can revise the outline, page descriptions, or generated slides, and can select a region of a slide to redraw. Per-page templates let each slide use its own template, with automatic style matching or manual binding.

Decks export as PPTX or PDF. Standard PPTX exports contain full-slide images. An editable PPTX export (Beta) extracts text and page elements, though complex layouts may not reproduce exactly. The project also documents narrated MP4 video export with AI voiceover and subtitles, and a CLI with agent skills.

Model Providers and Deployment

Users bring their own API keys. The README documents Gemini, OpenAI-compatible, Volcengine, Vertex AI, and LazyLLM provider formats, plus OpenAI OAuth sign-in to use Codex as a provider. Docker Compose with prebuilt images is the documented route, and a source install needs Python 3.10+, uv, Node.js 16+, and FFmpeg. The code is licensed under AGPL-3.0, with commercial closed-source use requiring authorization.

Current Status

The README lists version 0.9.0 release candidate 7 with a desktop build. The website also advertises a hosted beta waitlist.