A larger image is not new evidence
The ESRGAN slim 2× model estimates extra pixels from patterns learned during training. Edges may look clearer, but fine details may be invented or distorted. It cannot recover an original identity, document text, or an exact detail that the source never contained.
A practical workflow
Choose a photo, screenshot, illustration, or transparent image up to 50 MB, 16 megapixels, and 8192 pixels per side. Fast resize is the default and needs no model. Auto uses AI for images up to 0.5 MP and fast resizing for larger images. You can select AI enhancement for larger images: it processes overlapping tiles and can take several minutes. Fast resize enlarges pixels without inventing detail. Both modes export a 2× PNG and preserve transparency. Compare the result before saving. Animated images produce a single still frame. HEIC, RAW, TIFF and PSD need conversion to PNG or JPG first; other formats depend on your browser.
Before you export
Use only media you own or have permission to process. Keep the source and inspect the result. Desktop Chrome or Edge is a useful starting point; memory, codecs, and graphics support vary by device. A readable error means the current file or environment needs a different workflow, not that a result was created.
Can it fix every blurry photo?
No. Mild softness can look improved. Motion blur, heavy compression, tiny text, and damaged faces can remain unreliable. Keep the source file and decide by visual inspection.
Do my files leave this device?
No media is uploaded by these tools. Your browser downloads pages and models from the website host, and the media engine from jsDelivr. Hosting requests and contact inquiries are explained in the privacy policy.
Report a problem or suggest an improvement →
Read the current limits for this tool. Files are processed locally; ordinary website and dependency requests are explained in the privacy policy.