--- title: AI Face Restoration emoji: 🙂 colorFrom: green colorTo: gray sdk: static app_file: index.html pinned: false license: mit short_description: "Enlarge group photos and compare every face up close" tags: - face-restoration - face-enhancement - image-upscaling - super-resolution - image-restoration - image-to-image --- # Face Enhancer by the MegaPotato AI upscaler (megapotato.app) A free online face enhancer from the [MegaPotato AI upscaler](https://megapotato.app/?s=hf-face-enhancer-readme). Enlarge group photos, portraits and crowd shots ×2, ×4 or ×8, then pick any face on the photo and compare it before and after at the same spot and scale. In MegaPotato's [blind test of 18 AI upscalers](https://megapotato.app/blog/best-ai-image-upscalers-2026?s=hf-face-enhancer-readme), none of 67 small faces in group photos came out a different person; creative upscalers changed up to three in four. ## How to enhance the faces in a photo 1. Choose, drop or paste a PNG, JPEG or WebP photo. 2. Pick **×2**, **×4** or **×8**. The page shows the exact output size. 3. Click or tap the faces you want to check; each one gets a numbered box. You can do this before the run. 4. Press **Enhance faces**. Each box becomes a Before / After pair; drag a box to adjust it, switch the whole photo between Before and After. 5. To download the full-size result, upload the photo on [megapotato.app](https://megapotato.app/tool?s=hf-face-enhancer-readme) (new accounts get free chips). The boxes are placed by you. Nothing on the page detects or recognizes faces. The examples are real results. ## Free demo limits Enhance your own photos in this Space: **1 free run** without logging in (a new one 30 days after it is used), then **2 runs a day** when you log in to Hugging Face (per network address; the daily runs over a rolling 24 hours), with the same model and size limits as the product (up to 16 400 px per side). The examples work for everyone. ## How this Space works This Space is a static HTML page. Uploads go to a MegaPotato service on Cloudflare that holds the API key, applies the demo limits and passes the image to the MegaPotato API. The proprietary model runs on MegaPotato's GPUs; the model weights are not published. Images are handled under the MegaPotato [Privacy Policy](https://megapotato.app/privacy) and [Terms](https://megapotato.app/terms). Every image goes through an automated content check before processing, and images it blocks are refused. Uploads from visitors who are not logged in to Hugging Face are checked at a stricter level. The original is deleted once the run finishes and the result about an hour after the upload. The demo asks the Hugging Face page only whether you are logged in (not who you are). To apply the run limits and count visits, the demo service keeps a keyed hash of your network address, never the address itself (each demo counts separately): in the record of each run, with whether the page reported you as logged in, deleted 7 days after the run, or 90 days after it if its outcome is still unclear; in the record of the free run without a login, deleted 30 days after that run; and in a daily log of what the address did here (opened the page, started a run logged in or not, reached a limit or the login wall, failed the human check), deleted 120 days after that day ends (UTC). Daily totals without any address are deleted 120 or 400 days after their day ends. Each deletion happens at the next cleanup, which runs at least every 6 hours. Cloudflare processes your network address to run the service and the Turnstile check that helps limit automated abuse. This browser tab keeps the pass from that check (valid for an hour) and, while a run is open, its ids, size, scale and start time, so a reload can pick it up; the tab's storage is cleared when you close it. The same model is available through the [MegaPotato REST API](https://megapotato.app/developers?s=hf-face-enhancer-readme). ## Feedback Use this Space's Community tab to report errors or share feedback. Only share images you are comfortable making public. ## Credits and licensing Example photos: Cavan football team, 1914, Bain News Service, and Italian grocer, New York, 1943, by Marjory Collins (U.S. Office of War Information), both Library of Congress, no known restrictions, public domain; two 19th-century studio portraits from Wikimedia Commons, Edward Noel Hodges, 1877 (J. C. Turner, London) and Arthur König, 1870–1885 (Julius Lawitzky, Berlin; Kunstbibliothek Berlin), both cartes de visite, public domain, copyright expired. The demo inputs are small copies of the archive scans; the results are real MegaPotato outputs. Sources, each input's recipe and its license are listed in `examples/examples.json`. Font: MP Humanist, a subset of Lato under the SIL Open Font License (`assets/fonts/OFL.txt`). Demo code: MIT license; access to the proprietary model and API is governed by the MegaPotato service terms.