Enlarge a low-res portrait.
Sharpens hair, clothing, and facial edges, with optional face enhancement for small or compressed portraits.960x540 source, upscaled 4x to 3840x2160
Real-ESRGAN
Enlarge low-resolution images up to 4x while reducing blur, noise, and JPEG artifacts.
Try Real-ESRGAN
Drag the divider to compare. Upload your own image to run it.
From this.
To this.
960x540 source, upscaled 4x to 3840x2160
Before and after, on real images
Enlarge an old family scan.
Makes a small scan easier to print by sharpening edges and reducing noise, although scratches and severe damage may remain.960x540 source, upscaled 4x to 3840x2160
Clean up supplier product shots.
Clarifies product edges and fabric texture while preserving composition; inspect labels and logos before publishing.960x540 source, upscaled 4x to 3840x2160
Upscale AI artwork to 4K.
Reconstructs sharper edges and denser-looking texture in 1K artwork while keeping the original composition.960x540 source, upscaled 4x to 3840x2160
Scale images from 1x to 4x
Choose 1x, 2x, 3x, or 4x output, with 2x selected by default. BudgetPixel accepts images up to 2048x2048, so a maximum-size square input can produce an 8192x8192 result at 4x.
- •1x, 2x, 3x, or 4x output
- •2x default scale
- •2048x2048 maximum input
- •Up to 8192x8192 output
- •Dimensions scale proportionally
Reduces compression and noise
Real-ESRGAN is trained to handle mixed real-world degradation, including blur, noise, and JPEG compression. It can clean blocky areas and sharpen contours, but it does not reliably repair scratches or severe motion blur.
- •Reduces JPEG blocking
- •Cleans mild image noise
- •Sharpens edges and contours
- •Works well on small web images
- •Scratches may remain
- •Heavy blur can remain
Optional enhancement for small faces
Face enhancement can make eyes, hairlines, and facial contours look cleaner in low-resolution portraits. It may smooth skin or alter small identity traits, so review likeness carefully and never treat the result as evidence.
- •Optional face enhancement
- •Useful for small portraits
- •Can smooth skin texture
- •May alter eyes or fine features
- •Review likeness at 100% zoom
- •Not suitable for identification
Sharpens readable text and fine lines
The model can strengthen edges around signs, interface elements, and larger printed lettering. Tiny or unreadable text may be guessed or distorted, so verify every label, logo, and important number against the source.
- •Sharpens larger lettering
- •Cleans mild text compression
- •Strengthens interface edges
- •Thin lines may oversharpen
- •Cannot decode unreadable words
- •Labels need manual review
How it works
1
Upload your image
Choose an image up to 2048x2048. For the clearest result, start with the original file rather than a repeatedly saved or compressed copy.
2
Pick the scale and settings
Select 1x, 2x, 3x, or 4x; 2x is the default. Enable face enhancement for small portraits, then check whether the resulting likeness remains acceptable.
3
Download the result
Compare the original and enhanced image at 100% zoom. Download it as PNG or JPG once faces, text, logos, and fine lines look right.
Pricing for Real-ESRGAN
Runs on credits — no per-model surcharges, no surprise billing.
5credits
per generation
5 credits per image
Frequently asked questions
What is Real-ESRGAN and how does it work?▾
Real-ESRGAN is a GAN-based image super-resolution model designed for low-resolution and degraded real-world images. It learns to reduce common blur, noise, and compression artifacts while reconstructing sharper edges and plausible texture. It is not diffusion-based and does not require a text prompt.
Does Real-ESRGAN invent detail that was not in the original?▾
Real-ESRGAN uses learned visual patterns to reconstruct sharper edges and plausible texture. It cannot recover facts that were never captured, and it may guess or change tiny faces, text, logos, or fine patterns. Treat the output as a visual enhancement rather than a forensic or documentary source.
How much can Real-ESRGAN enlarge an image?▾
On BudgetPixel, you can select 1x, 2x, 3x, or 4x scaling, with 2x as the default. A 1280x720 image becomes 5120x2880 at 4x. The largest possible output is 8192x8192 when a 2048x2048 input is processed at 4x.
What is the maximum input size for Real-ESRGAN?▾
The maximum accepted input size on BudgetPixel is 2048x2048 pixels. If an image exceeds that limit, resize or crop it before uploading. Starting from the least-compressed available copy generally produces a cleaner result.
Does Real-ESRGAN work on faces?▾
Yes, optional face enhancement can improve the appearance of small, soft, or compressed faces. It can also smooth skin and change eyes, expressions, or other fine identity traits. Compare the result with the source and do not use it for identification or evidence.
Does Real-ESRGAN keep text readable?▾
It can sharpen text that is already partly readable and reduce compression around larger letters. It cannot reliably reconstruct characters that are too small or blurred to distinguish. Verify documents, signs, prices, labels, and logos against the original.
Can Real-ESRGAN fix blurry or compressed photos?▾
It works well on mild softness, JPEG blocking, ringing, and some visible noise. Severe defocus, motion blur, scratches, and missing areas may remain or become more noticeable after sharpening. The result depends heavily on how much usable structure exists in the source.
What output formats does Real-ESRGAN support?▾
BudgetPixel provides Real-ESRGAN results as PNG or JPG files. Choose PNG when you want to avoid another lossy compression pass. JPG is useful when a smaller file is more important.
Can I upscale AI-generated images with Real-ESRGAN?▾
Yes, it can enlarge AI artwork, illustrations, and rendered images up to 4x. It often makes edges and textures look denser, but may amplify malformed details or alter small faces and lettering. Inspect the output closely before printing or publishing.
How much does Real-ESRGAN cost?▾
Each enhancement uses BudgetPixel credits, with usage affected by the selected scale and processing requirements. You can review the current credit estimate before running the model. Processing time can vary with image size and settings.