BROWSER BETA · GENERATIVE INPAINTING
AI Object Eraser with Generative Fill.
Brush over an unwanted object. MI-GAN generates a plausible fill for the selected area, directly on your device — no photo upload to a Zhatpat server.
✦ No account◈ On-device AI↗ PNG export
STEP 01
JPG · PNG · WEBPUpload and mark
Paint over the object
Loading AI model…
First use downloads the MI-GAN model (about 28 MB). It may be cached by your browser. Processing speed depends on your device.
STEP 02
AI-GENERATED FILLReview your result
✧Your AI-edited photo appears hereUpload a photo and paint over an object
Download PNG ↓PNG · local preview resolution
BETTER RESULTS
Mark the whole object
- Cover the full object and a little space around its edges.
- Works best when the surrounding background gives the model useful context.
- Faces, text, repeating patterns and large complex areas may be reconstructed inaccurately.
This tool uses MI-GAN (Mobile Inpainting GAN) to generate plausible replacement pixels inside the marked region. It does not recover the exact hidden background. Review the result carefully, especially around faces, text, logos and detailed patterns. Input is resized to a maximum of 1 megapixel for device-memory safety. The image is processed in your browser and is not sent to a Zhatpat application server. The model and ONNX Runtime are downloaded from Hugging Face and jsDelivr.
Model: MI-GAN ONNX model and model card. Runtime: ONNX Runtime Web. Model licence: MIT; see attribution and licence.