[PDF2zh 2.0] Three Deployment Guides and Zotero Plugin Configuration
A beginner-friendly guide to configuring PDF2zh 2.0 and the Zotero plugin through Windows .exe, local uv, and Docker deployment, helping you build an efficient research-translation workflow.
Before we begin: make PDF-paper translation less painful#
Still overwhelmed by dense English PDF papers? PDF2zh 2.0 and its new BabelDOC engine improve layout fidelity and make the translation experience much smoother. This is a beginner-oriented guide for users with little programming experience. We cover three deployment paths:
Lazy path: launch the Windows .exe.
Developer path: create a local Python virtual environment with uv.
Stable path: deploy through Docker.
We also explain how to configure the Zotero-PDF2zh plugin so literature management and translation can work together.
Standalone research-PDF translation engine with CLI and Web UI
Preserves formulas/figures, supports bilingual or translated output, and works with multiple services
Python 3.10–3.12
Zotero-PDF2zh
Zotero plugin that calls a local PDF2zh service
One-click translation, two-column crop, and translated attachment management
Zotero 7
uv
Rust-based Python package and environment manager
Very fast dependency resolution and lightweight virtual environments
Win / macOS / Linux
Docker Desktop
Standard container platform
Reproducible, isolated, and easy to migrate
Windows/macOS/Linux
Compatibility warning:PDF2zh 2.0 (pdf2zh_next) is a new major version and is not compatible with 1.x services. If an old version is installed, uninstall it first with pip show pdf2zh and pip uninstall pdf2zh.
PDFs must contain selectable/copyable text. For scanned PDFs, enable the OCR workaround or auto-enable OCR workaround in the Web UI.
Author recommendation: use the .exe if you only want the GUI; use uv if you like Python and want local control; use Docker if you want stability and fewer environment surprises.
Prepare Python 3.10–3.12 for local deployment, Zotero 7 for plugin integration, Docker Desktop for container deployment, and an optional API key for higher-quality LLM translation.
Download a pdf2zh-<version>-with-assets-win64.zip package from the PDFMathTranslate-next releases. Prefer the with-assets version because fonts and models are already packaged and first-run failures are less likely. Unzip it, double-click pdf2zh.exe, and open http://[redacted-ip]:7860/ if the browser does not open automatically.
This route creates a clean local Python environment. First install Python and uv, create a project folder, create and activate a virtual environment, then install pdf2zh_next, pypdf, and flask. Use pdf2zh_next --gui for the Web UI, or run CLI translation directly with pdf2zh_next "path/to/paper.pdf".
For Zotero integration, install the Zotero-PDF2zh .xpi, prepare server.py, generate config.toml from the PDF2zh Web UI configuration, create a translated/ folder, and run uv run python server.py. In Zotero, set the engine to pdf2zh-next, point the config path to ./config.toml, and set the output directory to ./translated/ or leave it blank.
A one-click startup script can automate project-directory switching, virtual-environment activation, and server startup on Windows or macOS/Linux.
Docker packages the full PDF translation environment into a reproducible container. Install Docker Desktop, verify it with docker -v and docker info, then pull and run awwaawwa/pdfmathtranslate-next. Open http://[redacted-ip]:7860/ to use the Web UI. Use docker start, docker stop, docker logs, docker restart, and docker rm to manage the container.
For Zotero integration, create a Docker Compose project, download the official Dockerfile and docker-compose.yaml, create a persistent zotero-pdf2zh/translated directory, and mount config.toml plus translation outputs into the container. Replace the old base image with awwaawwa/pdfmathtranslate-next:latest when you want the newer 2.x engine.
Enable enhanced compatibility or use a stronger model
Custom prompts add token cost because the same prompt is attached to many translation chunks. If your provider supports KV Cache, the cost is much lower; otherwise disable custom prompts when cost matters.
RPS, QPS, and Zotero “thread count” all refer to the concurrency of requests sent to the upstream LLM. PDF preprocessing is mostly single-threaded, so speed optimization usually means choosing a faster API, improving network quality, and tuning concurrency within provider rate limits.
With PDF2zh, language should become a small stream to step over rather than a roadblock in research. I hope this guide helps you avoid pitfalls and translate papers more smoothly. Happy researching!
Use multimodal models such as GPT-4o with three reusable prompts to load PDFs, summarize article structure, generate deep-reading notes, and produce mind-map-ready references for research reading.
Explore Bohrium AI Search for second-level retrieval across 160 million papers, plus knowledge bases, notebooks, subscriptions, and other research-cloud functions that improve discovery and research efficiency.
Metaso AI Search, built by Shanghai Metaso Network Technology, combines its MetaLLM model with an ad-free structured AI teacher. It can generate step-by-step explanations, slides, mind maps, and multimodal outputs for study, research, and work.
Use custom prompts in PDFMathTranslate and Zotero for pdf2zh to improve academic translation quality while preserving structure, formatting, citations, and formulas.
This article explains how to download, install, and configure the Cherry Studio AI assistant desktop client on Windows, covering LLM service setup and common shortcuts so you can get started quickly and use it efficiently.
This article explains how to install and configure the pdf2zh plugin in Zotero, making it easy to translate PDF literature automatically and process two-column PDFs to improve academic reading efficiency.
This article explains common problems and practical solutions for PDFMathTranslate, covering image and table translation, model selection, installation configuration, and GUI usage to provide an efficient and stable PDF translation experience.
This article introduces an efficient polishing prompt based on large language models. It improves the writing experience, provides a quick-use option, and offers model-selection suggestions for more efficient content creation.
This article shares four frequently used browser extensions, including video speed control, dictionary and translation tools, and ad blockers, to help improve daily efficiency and browsing experience.