🔍 AI Image Upscaler
Enlarge a photo or an illustration ×2 or ×4 with the Real-ESRGAN neural network. The image is processed on your device: it is never uploaded.
Drop an image here, click to choose one or paste it (Ctrl+V)
JPG, PNG or WebP · best on small or low-resolution images
⚠️ The AI invents plausible detail, not real information. It cannot recover a face, a number plate or text that is not readable in the original. Do not use the result as evidence or for forensic purposes.
Support TinyToolWeb
Help us keep these tools free and ad-free!
About the AI Image Upscaler
Classic enlargement (bicubic) only spreads the existing pixels, so the image gets soft. This tool uses Real-ESRGAN, a neural network trained to rebuild edges and textures: it produces a sharper image ×2 or ×4 larger, directly in your browser with your graphics card (WebGPU) or your processor (WebAssembly).
Key Features
- ×2 and ×4: ×2 is computed at ×4 and then reduced, as in the official Real-ESRGAN tool.
- Two models: photos (realesr-general-x4v3) and illustrations, drawings or anime (realesr-animevideov3).
- Noise reduction: the official "denoise strength" blend between the standard and the "wdn" photo models.
- Comparison: before/after slider and two synchronised zooms on the detail of your choice.
- Transparency: the alpha channel of PNG and WebP images is kept and enlarged (bicubic).
How to Use
- Step 1: Drop, choose or paste an image.
- Step 2: Choose ×2 or ×4, the image type and the noise reduction, then click Upscale.
- Step 3: Compare with the slider and click on the preview to inspect a detail.
- Step 4: Download the result as PNG, JPEG or WebP (file name ending in -x2 or -x4).
Speed and Limits
- Speed: it depends on your hardware. With WebGPU, a small image takes a few seconds; on the processor (WASM), enlarging 512 × 512 px to 2048 × 2048 px can take a minute or more.
- First use: the AI engine (14 MB, or 27 MB for WebGPU) and the model (1 to 5 MB) are downloaded once, then kept in the browser cache.
- Tiles: the image is processed in overlapping tiles so memory stays limited; the progress bar shows the remaining time.
- Maximum size: 16.8 megapixels of output, i.e. 4096 × 4096 px (33.5 with WebGPU). For a large photo, choose ×2.
Common Use Cases
- Enlarging a small product photo, a logo or an old web image
- Getting a sharper print from a low-resolution picture
- Upscaling drawings, screenshots of illustrations and anime frames
- Cleaning noisy or heavily compressed JPEG images
Tips & Honest Limits
- Works best on photos and small, low-resolution images; a sharp 12 MP photo gains little.
- The details added by the AI are plausible guesses: faces, text and plates can look different from reality.
- Not suitable for evidence, identification or forensic analysis.
- If the result looks too smooth ("plastic"), lower the noise reduction.
Privacy & Security
The image never leaves your device: the model runs in your browser and nothing is sent to a server. The exported file is a new image, without the metadata (EXIF, GPS) of the original.