OSINT-mindset/lingolens

# lingolens Search in Google Lens in lingo! Tired of irrelevant results of reverse image search? Yeah, search results can be VERY different because of your language environment! Lingolens allows: - search images in Google Lens with specific languages and countries, excluding known results - visually select a region of the image to search (optional) - generate one HTML report with all the results - compare the target image with the result images via a sticky preview - a pretty user interface is supported! The full list of supported languages and countries is [here](https://developers.google.com/custom-search/docs/xml_results_appendices?hl=en#interfaceLanguages). ## How it works Google Lens used to return server-rendered HTML for the upload endpoint, so a plain `requests.post` was enough. That's no longer the case — Lens now requires a JavaScript-capable client and aggressively detects bots. So lingolens drives a real Chromium instance via [Playwright](https://playwright.dev/python/) with stealth patches and a persistent profile (cookies are reused, captcha state survives between runs). For each selected language, a fresh browser context with that `locale=` is opened. This makes the JS upload request honour the locale (`hl=ko`, `Accept-Language: ko-KR`, `navigator.language` etc.) and Google really returns different result sets across languages. ## User interface ## Report example Check example of search results: [report.html](report.html). ## Installation pip3 install -r requirements.txt playwright install chromium ## Usage ### As a browser-based tool streamlit run web_search.py Streamlit serves the UI at http://localhost:8501/. Pick languages (mandatory) and countries (optional), upload an image. Optionally toggle **Select a region of the image to search** — drag a box on the image, then **double-click inside the box** to apply the crop. Then press **Search in Google Lens with selected languages**. While the search runs you'll see a live status panel with per-language progress (`KO: 33 total on page, 15 new (rest already seen)`). After it finishes you get: - a **Download report** button (full HTML report), - a 3-column gallery of result thumbnails with `[LANG]` tags and links to source pages, - a sticky preview of the original/cropped image in the bottom-right, so you can visually compare while scrolling. #### First-run captcha The very first time Lens may show a captcha. If that happens, run once with: LINGOLENS_HEADLESS=0 streamlit run web_search.py This makes the Chromium window visible — solve the captcha manually. Cookies are stored in `~/.lingolens-profile`, so subsequent runs in default (headless) mode reuse them and shouldn't trigger captcha again. ### As CLI tool ./lingolens.py example.jpg You will get the `report.html` file in the same folder. Languages are read from `langs.txt`, one per line: ru en pl ### Environment variables | Variable | Default | What it does | | --- | --- | --- | | `LINGOLENS_HEADLESS` | `1` | `0` to show the Chromium window (useful for solving captcha) | | `LINGOLENS_PROFILE_DIR` | `~/.lingolens-profile` | Where Playwright stores its persistent profile (cookies, captcha state) | ## TODO - [x] Customization of language list for a search (simple config file) - [x] Language filter in a report - [x] Visual region selection (crop) - [x] Inline gallery and sticky original preview in the Streamlit UI - [ ] Standalone exe-file for Windows - [ ] Checkbox for switching to thumbnails instead of full images - [ ] Validation of lang-country combinations - [ ] Automatic captcha-solving fallback

标签：后端开发