--- title: "Image Eraser" description: "Free AI image eraser: erase part of an image — remove unwanted objects, people, text, or watermarks from any photo. Paint over it and the AI fills the gap naturally. Runs in your browser, no upload, no watermark, no signup. On first use it downloads an AI model once - 28 MB for Fast (MI-GAN) or 207 MB for Quality (LaMa) - then works offline." url: https://findutils.com/design/image-eraser/ category: design --- # Image Eraser Free AI image eraser: erase part of an image — remove unwanted objects, people, text, or watermarks from any photo. Paint over it and the AI fills the gap naturally. Runs in your browser, no upload, no watermark, no signup. On first use it downloads an AI model once - 28 MB for Fast (MI-GAN) or 207 MB for Quality (LaMa) - then works offline. **Use this tool:** [Image Eraser](https://findutils.com/design/image-eraser/) ## Programmatic access - Browser-only: this tool works on a file, the DOM, or a browser API and has no REST or MCP id. ## Why Use an AI Image Eraser? Traditional image editing requires manual clone-stamp work that takes time and skill. Our AI Image Eraser uses MI-GAN, a neural network specifically designed for image inpainting, to intelligently fill in removed areas by understanding the surrounding context. Unlike cloud-based erasers that upload your photos to external servers, this tool runs the AI model entirely in your browser. Your images never leave your device, making it safe for sensitive photos and proprietary content. The tool supports iterative erasing. After each erase operation, you can paint over additional areas and erase again, building up complex edits step by step. ## Frequently Asked Questions ### Are my images uploaded to any server? No. The AI model runs entirely in your browser using WebGPU or WebAssembly. Your images never leave your device. ### What is the AI model used? We use MI-GAN (Mobile Image GAN), a lightweight neural network designed for fast image inpainting. The model is only 28 MB and runs in real-time on modern browsers. ### Why does it download something the first time? The AI model (~28 MB) needs to be downloaded once to your browser. It is cached in your browser's storage, so subsequent visits load instantly. ### What browsers are supported? Chrome, Edge, and other Chromium-based browsers work best with WebGPU acceleration. Firefox and Safari work via WebAssembly fallback but may be slower. ### What is the maximum image size? Images are automatically scaled to a maximum of 2048 pixels on the longest side. The AI processes at 512x512 internally and blends the result back at your image's resolution. ### How does MI-GAN differ from LaMa, and which should I pick? MI-GAN (the Fast model) is a 28 MB GAN-based inpainting network optimized for mobile inference. It excels at small-to-medium masks (power lines, text, watermarks, small objects) and runs roughly 7x faster than LaMa. LaMa (the Quality model, 207 MB) is a Fourier-Convolutions inpainting network trained for higher fidelity on larger masks and complex textures. Choose LaMa for removing people from busy scenes, larger objects, or anything where MI-GAN produces visible blur. Switching is one click and re-runs against the same mask, so you can compare results without re-marking the area. ### Why does the result look blurry on large masks? Inpainting models trained on real-world photo datasets struggle when the mask exceeds about 30% of the image because there's not enough surrounding context to infer realistic content. Two mitigations: erase in passes (start with a smaller mask covering the most distracting parts, run the model, then re-mask any remaining artifacts) and switch from MI-GAN (Fast) to LaMa (Quality) for the large-mask case. If the result still looks wrong, the underlying scene may simply be too complex — a busy crowd or a heavily-textured background — and a manual clone-stamp pass in a desktop editor will do better. ### Does the AI model leak any data over the network after first download? No. After the model file (28 MB for MI-GAN or 207 MB for LaMa) is downloaded once and cached in your browser's IndexedDB, all inference happens locally using either WebGPU (Chrome, Edge) or WebAssembly (Firefox, Safari). The page does not phone home with image data, mask coordinates, or generated output. You can verify this by opening DevTools → Network and watching no requests fire during a normal erase operation. The first-load model download is the only network activity. ### What's the difference between this and Photoshop's Generative Fill? Photoshop Generative Fill (Adobe Firefly under the hood) is a diffusion model trained on Adobe Stock that can both REMOVE content (similar to this tool) and ADD new content from a text prompt. It produces higher-quality results on average and handles harder cases like people in crowds. The tradeoffs: it requires a Creative Cloud subscription, every operation uploads your image to Adobe's servers, generated content cannot be used commercially under the standard license, and it's overkill if you just need to remove a small distraction. The FindUtils Image Eraser is the right tool for fast, private, removal-only workflows. ### Can I use the AI-erased images commercially? Yes. The AI inference happens in your browser using open-source models (MI-GAN under MIT License, LaMa under Apache 2.0) and the output is yours to use however you want, including commercial work. The original copyright of the source image still applies — if you didn't have rights to the original photo, the erased version doesn't grant new rights. For Photoshop Generative Fill and Adobe Firefly the situation is different: Adobe's standard CC subscription license restricts commercial use of generated content, so this matters when picking the right tool for a paid project. ### How do I delete the 207 MB cached AI model? On this page, click Delete cached models. That clears the FindUtils IndexedDB store named findutils-models, which holds the MI-GAN (28 MB) and LaMa (207 MB) ONNX files. The next erase downloads the model you pick again. This does not remove Hugging Face files from the browser HTTP cache. ## Related Tools - [Image Resizer](https://findutils.com/convert/image-resizer/) - [Image Compressor](https://findutils.com/convert/image-compressor/) - [Image Color Extractor](https://findutils.com/design/image-color-extractor/) - [Image to Base64](https://findutils.com/convert/image-to-base64/) - [Favicon Generator](https://findutils.com/design/favicon-generator/) - [Social Media Image Resizer](https://findutils.com/convert/social-media-image-resizer/)