SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI officeLandPPT

LandPPT

LandPPT is a large language model-based presentation generation platform that can automatically convert topics or document content into editable professional PPTs. The platform integrates outlining, layout design, image matching, script writing, and export into a single AI workflow, supports input from multiple document types such as PDF, Word, Markdown, Excel, and PPT, and can optionally use deep research to supplement the latest information. It supports models such as OpenAI GPT, Anthropic Claude, Google Gemini, and Azure OpenAI, and is also compatible with OpenAI protocol interfaces such as DeepSeek, Moonshot, and Qwen, as well as local Ollama models, allowing models to be routed by role to control costs. The generation process produces HTML slides in parallel, supports templates and AI adaptive layouts, and provides polishing methods such as sidebar AI chat editing, image replacement, and visual references. In the delivery stage, it can export PDF, HTML, standard PPTX, image-based PPTX, scripts, and more, or generate a public sharing link containing narrated audio and subtitles. The platform provides global master templates and scenario templates, supports uploading reference PPTX files to extract layouts, template package management, and project-level AI adaptive templates. Deployment is flexible: locally, it defaults to SQLite plus in-memory cache, and can be tried with a single command; for production environments, docker compose is recommended to orchestrate Web, Worker, PostgreSQL, Valkey, MinIO, and the PPTX renderer. The project uses the Apache License 2.0.

Open website

About this tool

LandPPT is an intelligent presentation generation platform based on large language models, developed by sligter and open-sourced on GitHub. Its core goal is to integrate "write an outline -> design layouts -> match images -> write scripts -> export" into a single AI workflow, allowing users to automatically generate editable professional PPTs by entering a topic description or uploading documents, with support for scripts, narrated videos, and multi-format export.

  1. Purpose and Workflow LandPPT is aimed at users who need to quickly create presentations and can convert topics or documents such as PDF, Word, Markdown, Excel, and PPT into presentation content. The workflow is divided into five stages: input (topic description or document), planning (generate an editable outline, with optional deep research to supplement the latest information), generation (produce HTML slides in parallel, supporting templates and AI adaptive layouts), polishing (sidebar AI chat editing, image replacement, visual references), and delivery (export in multiple formats or generate a public sharing link, including narrated audio and subtitles). Users can first confirm requirements (topic, audience, page count, scenario template), then generate a structured outline and edit it visually, followed by optional content research and image configuration, and finally generate an HTML presentation based on the outline.

  2. Main Capabilities

  3. Model support: Supports OpenAI GPT, Anthropic Claude, Google Gemini, and Azure OpenAI, is compatible with OpenAI protocol interfaces such as DeepSeek, Moonshot, and Qwen, and supports local Ollama models; models can be routed by role (outline, slides, editing, templates, scripts) to facilitate cost control.

  4. Document parsing: Supports multi-format input including PDF, Word, Markdown, TXT, Excel, and PowerPoint, using MinerU and MarkItDown for high-quality parsing.

  5. Image matching: Image sources include the local image library, Pixabay, Unsplash, DALL·E, SiliconFlow, Pollinations, OpenAI, and Gemini. AI automatically matches and automatically handles size, format, and quality; the image service is disabled by default and must be enabled on demand via ENABLE_IMAGE_SERVICE.

  6. Template system: Provides global master templates and scenario templates (general, travel, education, etc.), supports uploading reference PPTX files to extract layouts, as well as project-level AI adaptive templates. Template packages includeLayouts such as cover, table of contents, body, image-text, and summary automatically match and fill SVG components after AI generates content. Supports adding built-in packages, importing template package files, creating from PPTX, and AI creation; PPTX import offers either "Visual Replication" or "Preserve Original Slide Background".

  7. Editing and Generation: The four-stage workflow consists of requirements confirmation, outline, task tracking, and PPT generation, supporting stage re-runs and recovery, visualized outlines, and one-click public sharing. Projects allow modifying structured content, switching layouts, replacing images, and locking pages; the decision model supports Jev / TypeSafe or OpenAI-compatible protocols.

  8. Scripts and Videos: Can generate speech scripts (DOCX / Markdown / PPT notes), per-page narration audio supports Edge-TTS or ComfyUI Qwen3-TD (reference audio can be uploaded), and can export narration videos (MP4, 1080p, 30/60fps, optional subtitles).

  9. Export and Sharing: Supports exporting PDF, HTML, standard PPTX, image-based PPTX, speech scripts, etc., with one-click public sharing; the share page supports narration audio and subtitles. Standard PPTX depends on APRYSE_LICENSE_KEY; image-based PPTX does not depend on Apryse and is better suited for complex CSS/icon fidelity.

  10. API and Integration: Provides API Key authentication, making it easy to integrate with CI, scripts, n8n, etc.; compatible with OpenAI's /v1/chat/completions, /v1/completions, /v1/models, and provides project-level export, sharing, and speech script interfaces, suitable for non-browser workflows.

III. Usage Local use defaults to SQLite plus in-memory cache, allowing trial with a single command, with no need for PostgreSQL / Valkey; it can also run via a single Docker container or Docker Compose. For production environments, docker compose orchestration of Web, Worker, PostgreSQL, Valkey, MinIO, and the PPTX renderer is recommended. Before running, you need to create and configure .env (including at least the AI API Key). In production environments, administrator auto-initialization is disabled by default, and the first deployment requires explicitly enabling LANDPPT_BOOTSTRAP_ADMIN_ENABLED. Change SECRET_KEY and the administrator password; do not use the default admin / admin123; configure strong random LANDPPT_API_KEY / LANDPPT_API_KEYS. Narration video export requires ffmpeg. By default, migrations are automatically detected and executed on startup, which can be disabled via LANDPPT_AUTO_MIGRATE_ON_STARTUP=false; when multiple nodes share the same database, it is recommended to disable automatic migration and instead run a one-time migration job separately. Upgrading to the template package feature requires migration 019. The frontend source code is located in frontend/ and is compiled via Vite to src/landppt/web/static/antd/; when modifying the frontend, you can run npm run dev --prefix frontend for continuous compilation. Before releasing the Python wheel/sdist, the frontend build must be executed first. The access address defaults to http://localhost:8000。

IV. Applicable Scenarios LandPPT is suitable for individuals and teams that need to quickly turn documents or topics into presentations, such as work reports, teaching courseware, travel introductions, product demos, etc.; its template packs and quick mode are suitable for batch-generating PPTs in a unified style; the API interface is suitable for integration into automated processes or non-browser workflows; support for local Ollama models is suitable for local deployment scenarios with data privacy requirements.

V. Limitations and Notes

  1. Standard PPTX export depends on third-party components such as Apryse, and compliance must be confirmed independently; image-based PPTX does not depend on Apryse and is better suited for preserving complex CSS/icons.
  2. The "visual replication" of PPTX uses AI to reconstruct an editable layout by referencing images of the original pages, and may adjust crowded layouts or simplify decorations; "retain the original base image" keeps complex graphics as a fixed background and only replaces confirmed content slots. Both methods require extracted editable text, and complete OCR of scanned images or embedded charts is not guaranteed. Currently .pptx / .potx / .ppsx are supported, with a maximum file size of 50 MB and a maximum of 40 pages selectable each time.
  3. If images still point to localhost after a reverse proxy, the correct BASE_URL must be set in the Web "Application Configuration."
  4. Explainer videos require ffmpeg available on the local machine or in the container.
  5. The image service is disabled by default, and ENABLE_IMAGE_SERVICE must be enabled as needed and the corresponding API Key configured.
  6. The research feature requires configuring TAVILY_API_KEY or deploying SearXNG.
  7. If the template list or template pack interface returns 500 after an upgrade, confirm that migration 019 has been completed; an old database missing fields such as template_kind will cause query failures.
  8. When installing with Docker or from source, the inference side can connect to Ollama without an external large model API, but capabilities such as research and web image sourcing may still require external network access.
  9. Quick mode does not require separately configuring Jev, and still uses the configured AI model to generate content; layout selection can use the existing model or rules; if a dedicated decision model is needed, it can be set in "System Configuration → Decision Model Configuration."
  10. Parallel generation only changes the scheduling order; the single-page generation logic and quality remain unchanged, but it is subject to the rate limits of each provider.

The project uses the Apache License 2.0. The project homepage is https://github.com/sligter/LandPPT , issue feedback and the discussion area can be found on the corresponding GitHub page, and the contact email is ai@yydsapp.com.

Related tools

Browse category