How do I enhance a picture's resolution?

Drop the picture on this page and leave the scale on 4×. The Real-ESRGAN model runs on your own device and writes a file with four times the width and four times the height — 16× the pixel count — reconstructing edges and texture as it goes instead of just stretching what was there. It is free, needs no account, and the download keeps the original file format.

How much can the resolution actually increase?

Up to 4× per side, which is the native factor of the realesr-general-x4v3 network; the 2× option is derived from the same pass. At 4× the model reads up to 2.5 megapixels of input on a device with WebGPU — roughly a 1920×1280 source — because the assembled result has to stay under the 40-megapixel output ceiling. Feed it something bigger and it is read slightly smaller first; the tool prints the exact input and output dimensions under the preview.

Can I convert a picture to 4K resolution?

Yes, and the arithmetic is easy to hit: 4K UHD is 3840×2160, so any source close to 960×540 lands on 4K exactly at 4×. A 1280×720 frame overshoots to 5120×2880, which you can crop or let your player downscale — a downscale from above 4K looks better than an upscale to it. For a source already near 1920×1080, the 4× pass produces 8K (7680×4320).

Does AI upscaling add real detail or just make the picture bigger?

It adds plausible detail, which is not the same as recovering the original. Real-ESRGAN was trained on millions of image pairs, so it knows what a sharp edge, a brick texture or a strand of hair should look like at 4× and draws that instead of interpolating a blur. What it cannot do is invent information that was never captured: text too small to read stays unreadable, and small faces can come back subtly restyled. Compared with the bicubic upscale your image viewer does, the difference on edges and texture is not subtle.

Is there a resolution or file-size limit?

The output ceiling is 40 megapixels and the batch limit is 20 pictures; there is no file-size limit and no daily quota, because there is no server involved. On CPU-only devices the AI pass reads at most 0.7 megapixels (about 1024×683) to keep a run down to a minute or so — larger sources get downscaled before the pass, which still gives you a net resolution gain at 4×.

Will a higher-resolution file print better?

Usually yes, because print quality is pixels per inch, not DPI metadata. At the 300 PPI print shops ask for, a 640×480 photo covers about 2×1.6 inches; the 2560×1920 result covers 8.5×6.4 inches. Upscaling does not change the metadata number your editor displays — it changes how many real pixels exist, which is the part that matters on paper.

How long does a 4× pass take?

A few seconds per picture where WebGPU is available (most desktop Chrome, Edge and recent Firefox), since the model runs on the graphics card. Without GPU acceleration it falls back to single-threaded WebAssembly and takes noticeably longer — a tile-by-tile progress bar and a Cancel button keep it manageable. iPhone and iPad deliberately use the slower CPU path because WebKit's WebGPU stack still mishandles the runtime.

Does my picture get uploaded to increase its resolution?

No. There is no upload endpoint on this site at all — decoding, the AI pass and re-encoding happen inside your browser tab, so no pixel leaves your device. The only traffic is in the other direction: a one-time 4.9 MB model download on first use, cached afterwards. It also means no upload wait, which matters most for exactly the large files other upscalers cap.

Free picture resolution enhancer

Enhance picture resolution 4× larger, not 4× blurrier

Stretching a small picture in an image viewer gives you the same detail spread over more pixels. This page runs Real-ESRGAN super-resolution instead, with the scale preset to 4×: it redraws edges and texture at the new size, so a 960×540 image comes out at exactly 4K and a 1080p screenshot reaches 8K. Free, no account, and the whole pass runs on your own device.

Drop a picture below — the upscale starts by itself, and the line under the preview tells you the output resolution in pixels.

Always free
No account
4× preset
16× more pixels
Up to 40 MP output
Runs on your device

Drop a picture here — AI enhancement starts on its own

JPG, PNG or WebP — up to 20 pictures at once

How to enhance picture resolution

Three steps, free, and the picture never leaves your browser.

Add the low-resolution picture
Drag a JPG, PNG or WebP onto the drop zone, or tap Choose pictures. Small files are the point here — thumbnails, chat downloads, old web exports and tight crops all work, and up to 20 go in one batch.
Leave the scale on 4× larger
This page opens with the AI scale preset to 4×, so the Real-ESRGAN pass starts on its own as soon as the picture loads. The line under the preview shows the exact arithmetic — source 1280×720 px → output 5120×2880 px — so you can see the new resolution before you commit. Switch to 2× larger if 4× overshoots what you need.
Check the detail, then download
Drag the split handle across the preview to compare the upscale against the original at the same display size, nudge Sharpness or Auto levels if the result needs it, and press Download enhanced — or Download all for the batch.

What 4× gets you, in pixels

Both sides multiply by four, so the pixel count multiplies by sixteen. These are real outputs — every source below fits inside the input cap that applies at 4×.

Source	Output at 4×	Typical case
480 × 360 px	1920 × 1440 px	Old email attachment → full-screen
640 × 480 px	2560 × 1920 px	VGA snapshot → large print
960 × 540 px	3840 × 2160 px	Small web image → exactly 4K UHD
1280 × 720 px	5120 × 2880 px	720p frame → 5K
1920 × 1080 px	7680 × 4320 px	1080p screenshot → 8K

At 4× the model reads up to 2.5 megapixels on a WebGPU device (0.7 MP on CPU-only ones) so the assembled result stays under the 40-megapixel ceiling. Anything larger is read at a reduced size first, and the preview caption names both numbers before you download.

Questions about increasing picture resolution

How do I enhance a picture's resolution?: Drop the picture on this page and leave the scale on 4×. The Real-ESRGAN model runs on your own device and writes a file with four times the width and four times the height — 16× the pixel count — reconstructing edges and texture as it goes instead of just stretching what was there. It is free, needs no account, and the download keeps the original file format.
How much can the resolution actually increase?: Up to 4× per side, which is the native factor of the realesr-general-x4v3 network; the 2× option is derived from the same pass. At 4× the model reads up to 2.5 megapixels of input on a device with WebGPU — roughly a 1920×1280 source — because the assembled result has to stay under the 40-megapixel output ceiling. Feed it something bigger and it is read slightly smaller first; the tool prints the exact input and output dimensions under the preview.
Can I convert a picture to 4K resolution?: Yes, and the arithmetic is easy to hit: 4K UHD is 3840×2160, so any source close to 960×540 lands on 4K exactly at 4×. A 1280×720 frame overshoots to 5120×2880, which you can crop or let your player downscale — a downscale from above 4K looks better than an upscale to it. For a source already near 1920×1080, the 4× pass produces 8K (7680×4320).
Does AI upscaling add real detail or just make the picture bigger?: It adds plausible detail, which is not the same as recovering the original. Real-ESRGAN was trained on millions of image pairs, so it knows what a sharp edge, a brick texture or a strand of hair should look like at 4× and draws that instead of interpolating a blur. What it cannot do is invent information that was never captured: text too small to read stays unreadable, and small faces can come back subtly restyled. Compared with the bicubic upscale your image viewer does, the difference on edges and texture is not subtle.
Is there a resolution or file-size limit?: The output ceiling is 40 megapixels and the batch limit is 20 pictures; there is no file-size limit and no daily quota, because there is no server involved. On CPU-only devices the AI pass reads at most 0.7 megapixels (about 1024×683) to keep a run down to a minute or so — larger sources get downscaled before the pass, which still gives you a net resolution gain at 4×.
Will a higher-resolution file print better?: Usually yes, because print quality is pixels per inch, not DPI metadata. At the 300 PPI print shops ask for, a 640×480 photo covers about 2×1.6 inches; the 2560×1920 result covers 8.5×6.4 inches. Upscaling does not change the metadata number your editor displays — it changes how many real pixels exist, which is the part that matters on paper.
How long does a 4× pass take?: A few seconds per picture where WebGPU is available (most desktop Chrome, Edge and recent Firefox), since the model runs on the graphics card. Without GPU acceleration it falls back to single-threaded WebAssembly and takes noticeably longer — a tile-by-tile progress bar and a Cancel button keep it manageable. iPhone and iPad deliberately use the slower CPU path because WebKit's WebGPU stack still mishandles the runtime.
Does my picture get uploaded to increase its resolution?: No. There is no upload endpoint on this site at all — decoding, the AI pass and re-encoding happen inside your browser tab, so no pixel leaves your device. The only traffic is in the other direction: a one-time 4.9 MB model download on first use, cached afterwards. It also means no upload wait, which matters most for exactly the large files other upscalers cap.

Why resizing and upscaling are not the same thing

Every image viewer can make a picture bigger. It does so by interpolation: to invent the pixel that belongs between two known ones, it averages them. Repeat that fifteen times per original pixel and you have a 4× file that carries exactly as much information as the small one did, with softer edges to show for it. That is why an enlarged photo looks worse the closer you get to it.

A super-resolution network guesses instead of averaging. Real-ESRGAN saw a very large number of high-resolution photos next to deliberately degraded copies of themselves, and learned the mapping backwards — what a crisp eyelash, a roof tile or a letterform looks like when the small version has only a few pixels of it. At inference time it applies that knowledge, which is why edges come back defined rather than smeared.

The honest caveat is that guessing is guessing. The model produces a plausible high-resolution image, not the photograph a better camera would have taken, so treat the output as a better-looking rendition rather than as evidence. Detail that was destroyed — a licence plate three pixels tall, a face in the background of a thumbnail — stays destroyed, and forcing 4× on such a source mostly enlarges the guesswork.

Resolution is also only one of the two things that make a picture look bad. If yours has enough pixels but suffers from compression blocks, noise or flat tone, the enhance photo quality page runs the same model at 1× and cleans it up at its original dimensions. If you are not sure which problem you have, start on the main picture enhancer at 2×, or read how each stage of the pipeline works on the about page.

No upload, no size cap

The upscale runs in your browser through WebGPU or WebAssembly, so there is no server to receive your picture, no queue, no daily allowance and no file-size limit imposed by an upload form. The one-time 4.9 MB model download on first use is the only network request the enhancer makes — after that it keeps working with the network switched off.

Enhance picture resolution 4× larger, not 4× blurrier

How to enhance picture resolution

Add the low-resolution picture

Leave the scale on 4× larger

Check the detail, then download

What 4× gets you, in pixels

Questions about increasing picture resolution

Why resizing and upscaling are not the same thing

No upload, no size cap