Local image utility · zero image uploads
Remove the visible Gemini logo from the corner of an image you own or are allowed to edit. The image stays on your device: decoding, watermark detection, pixel restoration, preview, and PNG export all happen inside this browser tab.
or choose a file with the button below
PNG, JPEG, or WebP · up to 25 MB · 18 megapixels · 8192 px per side
Same photo after 96px corner cleanup.
Bottom-right corner, shown 1:1.
A small tool with strict boundaries
A useful cleanup tool should be fast, private, and honest about what it changes. This page is designed for people preparing their own Gemini-generated images for presentations, mockups, mood boards, product drafts, and other authorized work. It does not send an image to a hidden queue or ask for an account before showing the result.
01 · Private by construction
Your selected file is decoded into a temporary canvas in this tab. Detection and restoration operate on that local pixel buffer. The preview uses a browser-only object URL, and the downloadable PNG is created from the restored canvas. Reloading or resetting clears the working state; there is no image upload endpoint in this flow.
02 · Region-aware
Gemini output sizes and visible logo layouts have changed over time. The detector compares calibrated alpha profiles and known corner positions for small 36px, standard 48px, and large 96px variants. Restoration is attempted only when the visual evidence passes the detector’s safeguards, reducing needless edits to ordinary corner details.
03 · Reversible workflow
The original file remains untouched. After a supported mark is detected, the page shows the restored result beside the original dimensions and reports the affected region. You decide whether to download. The export is a new PNG, so your source file, filename, and original bytes remain available exactly as they were.
Three local stages
The process reverses a known semi-transparent overlay rather than smearing, cropping, or generatively inventing a large patch. That narrow approach preserves more of the original image and creates a result you can inspect immediately.
Choose the highest-quality Gemini image available. PNG is ideal, although JPEG and WebP are accepted. The browser validates the declared type, file signature, byte size, decoded dimensions, and total pixel count before the processing button becomes available.
The detector evaluates expected bottom-right locations along with alternate margins used by known Gemini image sizes. It compares the local luminance structure against calibrated watermark alpha maps. If confidence is insufficient, the operation stops and the image data is left unchanged.
For a confirmed region, the algorithm estimates the underlying pixels by reversing the watermark’s alpha blend and evaluating conservative candidates for residual edges or clipping. Only the local region is written into the output buffer. The browser then creates a PNG for review and download without adding a JPEG or WebP encoding pass.
Clear scope
The words “watermark” and “provenance” are often treated as interchangeable, but they describe different layers. This tool is intentionally limited to a visible pixel pattern. It neither claims nor attempts to make an AI-generated image untraceable.
This tool targets the visible corner mark only. It does not remove SynthID, Content Credentials, provenance metadata, or any invisible signal.
A PNG export may represent metadata differently from the source format because the browser creates a new file, but that is not a promise that provenance systems or invisible identifiers are removed. If your workflow requires disclosure, attribution, or a specific metadata policy, follow the relevant platform and organizational requirements.
Compatibility notes
Use the direct Gemini download when possible. Screenshots, social-media recompression, aggressive JPEG artifacts, color filters, crops, and rescaling can alter the watermark profile or move it away from a known anchor. The detector may then decline to edit, which is safer than applying a destructive guess. A decline does not mean the file is invalid; it means the visible mark was not confirmed with enough confidence.
The current browser limits cover standard Gemini exports through roughly 4K-class pixel counts while protecting the tab from excessive memory use. Very large panoramic files may exceed 18 megapixels even when each side is under 8192 pixels. Process those files at their original supported size or use an audited local desktop workflow with sufficient memory.
Output is always PNG to avoid adding a second lossy compression pass. Transparent source pixels are preserved by the canvas pipeline. Color-profile handling can vary between browsers, so visually critical work should be checked in the browser and again in the application where the image will be used.
Responsible use
Use this utility for images you created, commissioned, licensed, or otherwise have permission to modify. A technical ability to change visible pixels does not grant copyright, publicity, trademark, or contractual rights. Keep records of prompts, licenses, approvals, and source files when your project requires them.
Do not use the tool to misrepresent authorship, conceal a required disclosure, bypass a platform policy, or remove a third party’s ownership mark. ImageGen2 is not affiliated with or endorsed by Google or Gemini. Product names are used only to describe compatibility with a visible output pattern.
The implementation is based on the MIT-licensed open-source research published by GargantuaX and contributors in the gemini-watermark-remover project, package version 1.0.41. The runtime is restricted to its image-data entry point; video processing, machine-learning models, WASM runtimes, and remote services are not used by this page.
Questions before you process
These answers describe the actual browser workflow and its limitations. They are intentionally specific so you can decide whether the tool fits your image and your usage rights.
No. The selected image is read into a canvas in your browser, processed locally, and exported from that canvas. The flow does not call an upload, storage, presign, or image-processing API. Normal page assets such as JavaScript may still be loaded from this website.
The detector includes calibrated profiles for 36px, 48px, and 96px visible marks, plus known standard and alternate corner margins. Results are strongest on original Gemini downloads. Resized, cropped, filtered, or recompressed copies may not match safely and can be rejected without editing.
No. The algorithm changes a small visible pixel region only. It does not target Google SynthID, invisible watermarking, C2PA or Content Credentials, platform records, or other provenance mechanisms. Do not use the PNG export as evidence that those systems are absent.
The workbench reports that no confirmed visible mark was found and does not write a processed result. Your source File object and source preview stay unchanged. This fail-closed behavior avoids modifying ordinary artwork merely because it has detail near a corner.
PNG avoids another JPEG or WebP quality pass and supports transparency. The tradeoff is a potentially larger file. You can convert the downloaded PNG later if a destination requires a different format, but inspect the converted result for compression artifacts.
The local visible-mark workflow on this page is free and requires no sign-in or credits. Browser safety limits still apply. ImageGen2’s separate AI generation and editing features may have their own pricing and account requirements, but they are not invoked by this utility.
Ready when your image is
Return to the workbench, use the original-size authorized image, and review the pixels before exporting. For broader creative changes, continue to the AI image editor after your local cleanup is complete.
Independent compatibility tool. Not affiliated with Google. Gemini is a trademark of Google LLC. Visible-mark removal does not remove or disable invisible provenance systems. Open-source license notice