Run Real-ESRGAN directly in your browser. Upscale images 2x, 3x, or 4x with AI super resolution and optional GFPGAN face enhancement. No install, no login, 100% free.
Upload an image to enhance its resolution
Upload an image and click upscale to enhance
Real-ESRGAN (Real-Enhanced Super-Resolution Generative Adversarial Network) is an open-source AI model developed by Xinntao for single image super resolution. Unlike traditional upscaling methods that simply interpolate between pixels, Real-ESRGAN uses a deep neural network trained on paired low-resolution and high-resolution images to generate realistic texture detail that does not exist in the original.
The model is trained using a second-order degradation pipeline that simulates real-world image degradation including blur, noise, JPEG compression artifacts, and resolution loss. This makes Real-ESRGAN significantly more robust than earlier models like ESRGAN, which were trained only on clean downscaled images.
Real-ESRGAN Performance Data
Real-ESRGAN achieves a PSNR of 31.5 dB on the RealSR benchmark — 2.1 dB higher than traditional bicubic interpolation (29.4 dB). The model upscales a 1920x1080 image to 7680x4320 (8K) in approximately 30 seconds. The default 4x model uses 16.7M parameters with a RRDB (Residual-in-Residual Dense Block) architecture.
GFPGAN (Generative Facial Prior GAN) is a companion model that specializes in face restoration. When enabled alongside Real-ESRGAN, it detects faces in the upscaled output and applies targeted enhancement to eyes, skin texture, hair, and facial features — producing sharper, more natural-looking portraits than super resolution alone.
Exact pixel dimensions for each input resolution and scale factor
| Input Resolution | 2x (4K) | 3x (6K) | 4x (8K) |
|---|---|---|---|
| 720p (1280x720) | 2560x1440 | 3840x2160 | 5120x2880 |
| 1080p (1920x1080) | 3840x2160 | 5760x3240 | 7680x4320 |
| 1440p (2560x1440) | 5120x2880 | 7680x4320 | 10240x5760 |
Maximum recommended input resolution is 1440p (2560x1440). Larger inputs may exceed processing limits. Output format is high-quality PNG.
Three steps to upscale any image with AI super resolution
Drop your image into the Real-ESRGAN upscaler. Supports JPG, PNG, and WebP formats up to 4.5MB.
Select 2x for 4K, 3x for 6K, or 4x for 8K output. Toggle GFPGAN face enhancement for portrait photos.
Get your high-resolution PNG output in 30-60 seconds. No watermark, ready to use immediately.
How Real-ESRGAN compares to popular image upscaling alternatives
Real-ESRGAN generates new texture detail using a neural network; bicubic interpolation just smooths between existing pixels. Real-ESRGAN achieves a 2.1 dB PSNR improvement over bicubic on the RealSR benchmark (31.5 dB vs 29.4 dB).
Real-ESRGAN handles real-world degradation types (noise, blur, JPEG compression artifacts). Waifu2x is optimized specifically for anime-style art and may produce artifacts on photographs with complex textures.
Similar output quality for most use cases, but Real-ESRGAN is 100% free and open-source. Topaz Gigapixel AI costs $99.99/year and requires a desktop installation with GPU.
Real-ESRGAN preserves your original image content pixel-perfectly. AI generators like DALL-E may alter details, add hallucinated elements, or change the composition during upscaling.
GFPGAN adds AI face restoration on top of Real-ESRGAN super resolution
Sharpens eyes, skin texture, and facial features for natural-looking results at any resolution.
Enhances multiple faces simultaneously, restoring clarity to each person in the frame.
Restores degraded facial details in scanned or low-resolution vintage photographs.
Improves facial clarity for official documents, visa applications, and identification cards.
When to disable GFPGAN
Disable GFPGAN for landscapes, architecture, anime/manga, text documents, product photography, or any image without human faces. The face detection model may produce unwanted artifacts on non-face regions that resemble facial patterns.
Yes, 100% free. No login, no watermark, no software installation needed. Upload any image, choose your scale factor (2x, 3x, or 4x), and download the upscaled result. Our Real-ESRGAN web tool runs entirely in the cloud.
Real-ESRGAN supports up to 4x upscaling, producing 8K output from 1080p input. A 1920x1080 image becomes 7680x4320 at 4x. You can also choose 2x (4K: 3840x2160) or 3x (6K: 5760x3240) depending on your needs.
GFPGAN (Generative Facial Prior GAN) is a face-specific AI restoration model. Enable it when upscaling portraits, group photos, or any image with faces. Disable it for landscapes, anime, text documents, or product photography to avoid unwanted face-detection artifacts.
Most images process in 30-60 seconds depending on input size and scale factor. A 1080p image at 4x takes approximately 30 seconds. Larger inputs (1440p) or higher scale factors may take up to 60 seconds.
For real-world photographs, yes. Real-ESRGAN is trained on diverse degradation types and achieves 31.5 dB PSNR on RealSR. Waifu2x is still an excellent choice for anime-only content, but Real-ESRGAN outperforms it on photos with noise, blur, and compression artifacts.
No. Our Real-ESRGAN tool runs entirely online in your browser. No Python environment, no GPU, no command-line setup. Works on any device with a modern browser — desktop, tablet, or mobile.
Upscale images to 8K (7680x4320) resolution. 33.2 megapixels for large prints and displays.
Try FreeAI face restoration powered by GFPGAN. Restore blurry faces, enhance eyes and skin texture.
Try FreeGeneral-purpose HD upscaler to 4K/6K/8K. Fast, free, no login required.
Try FreeReal-ESRGAN optimized for anime, manga, and illustration upscaling.
Try FreeAI lossless image enlargement. 免费在线图片放大工具，支持4K/6K/8K。
Try FreeUpscale photos to 4K, 6K, or 8K with optional GFPGAN face enhancement.
Try FreeUpscale any image to 4K, 6K, or 8K. No install, no signup.
Upscale Image Free