bingbongbang01/pdf_ai_translater

PDF AI Translator is a Python tool that translates PDF documents while keeping the original layout intact. Instead of producing a plain reflowed text file, it swaps out the text in place, preserving the original font size, position, tables, color, and boldness, so the translated PDF looks like the source document with a different language on it. It comes as both a command line tool and a Windows desktop GUI app. Translation can be powered by Claude, Gemini, or GPT through your own API keys. You can register keys for several of these providers at once, and if one key runs out of quota the tool automatically switches to the next available key or provider so a long translation job does not stop. For people with an AMD Ryzen AI chip that has an NPU, the tool can also fall back to translating locally through the Lemonade Server, which means a document can be fully translated without any internet connection or API key at all. The tool pays special attention to tables and multi column layouts, detecting when separate pieces of text sit close together and placing each translated piece back in its original spot so cells and columns do not break apart. If a translation job gets interrupted partway through, you can feed the partial output file back in and it automatically detects which page ranges were not translated and continues from there. There is also support for a glossary file so you can force specific translations for terms specific to your organization or project. The GUI version lets you pick your input PDF, enter API keys that are saved for next time, choose source and target languages with automatic detection, and watch a progress bar with an estimated time remaining. It can also install its own prerequisites automatically. Requirements are Python 3.12, and Windows 10 or 11 if you want the GUI, though the command line version also runs on Linux and macOS. Known limitations include no support for scanned image-only PDFs without adding OCR text first, lower quality from smaller local NPU models compared to cloud models, and occasional misalignment on very complex layouts like rotated text or magazine-style columns. The project is released under a Creative Commons Attribution NonCommercial license, so commercial use is not permitted.