Image Upscaler
Bigger, but not blurrier. A classical super-resolution pipeline reads edge direction to interpolate, then refines the result against the original.
Click or drop an image here
JPG · PNG · WEBP · Up to 50MB · Multiple files supported · Paste with Ctrl+V
Upload an image to see the upscaling options.
Scale
Image type
Output format
Done
Batch
사용 가이드
What it does
Enlarges small images by 2x, 3x or 4x while suppressing the blurriness and staircase artifacts that plain scaling produces. The image passes through DCCI (edge-directed interpolation), iterative back-projection that shrinks the result back down and compares it against the original, and an unsharp mask applied selectively to edges. Everything runs inside your browser — your image is never sent to a server.
How to use
- Upload an image: Drag and drop onto the drop zone, or click to choose a file. Select several files at once to process them as a batch.
- Pick a scale: Choose 2x, 3x or 4x. The resulting dimensions are shown alongside.
- Pick a preset: Choose photo, illustration or document to match your image type.
- Click Upscale to start processing.
- Compare: Drag the slider left and right to compare a plain enlargement against the result. The difference is clearest at 100% view.
- Download: Save as PNG, JPG or WebP.
Presets
- Photo: Portraits, landscapes and other ordinary photos. Preserves texture while cleaning up edges.
- Illustration / line art: Drawings, logos and line work with crisp color boundaries. Applies directional decisions more aggressively to reduce staircasing along lines.
- Document / text: Scans and screenshots where text is the subject. Weighted toward making strokes legible.
How it works
This tool does not use a trained AI model. It runs on classical image-processing algorithms alone, in this order:
- JPEG deblocking: Softens the 8x8 compression block artifacts that would otherwise be magnified along with the image.
- Linear-light conversion: All math is done in a gamma-undone linear color space to avoid brightness distortion.
- DCCI directional interpolation: For each pixel, measures variation along both diagonals and interpolates along the direction that does not cross an edge.
- Iterative back-projection: Shrinks the enlarged result back to the original size, compares it with the original, and feeds the difference back so the result converges on something that "becomes the original when shrunk".
- Anti-ringing and adaptive sharpening: Suppresses ghosting around edges and sharpens only edges, leaving noise in flat areas untouched.
On devices that support WebGL2 the work runs on the GPU; otherwise it falls back to a CPU path automatically. Both paths produce the same result.
Limitations and notes
- It cannot invent detail that is not in the original. Classical algorithms only make the most of what is there — unlike trained AI super-resolution, they will not draw in hair strands or letter strokes that were never captured. Enlarging a badly degraded low-resolution original by 4x will show this limit clearly.
- Up to 50MB per file, and up to 20 images in a batch.
- Browsers cap how large a canvas can be, so if the result would be too large you will be asked to lower the scale. This cap is especially low on iOS.
- On the CPU path, high-scale enlargement of a large image can take tens of seconds. Your browser stays responsive while it runs.
Frequently asked questions
How is this different from an AI upscaler?
AI super-resolution invents plausible detail that was never in the original, based on patterns it learned. This tool generates nothing — it enlarges using only information actually present in your image. Nothing fabricated is introduced, but on badly degraded originals the improvement is correspondingly limited.
Is my image uploaded to a server?
No. Decoding, enlarging and encoding all happen inside your browser. The result file is only transmitted if you press the cloud save button yourself.
The result doesn't look much different from the original.
If the comparison view is scaled down, the difference is hard to see. Switch to 100% view and drag the slider left and right. Edges and text show the largest difference.
What scale should I use?
If your original is sharp, 2x is the most reliable. Use 3x or 4x when you specifically need the physical dimensions, such as for print or a large display; if image quality itself is the goal, 2x is the better choice.