Increase resolution by 2×, 4×, or 8× with Real-ESRGAN — sharper detail, no blur
or drop here · JPG, PNG, WebP — up to 10 MB
Upload a file to start
4× and 8× run for 15-60s but recover meaningful detail. 2× is fast and good for moderate sharpening.
Powered by Real-ESRGAN (state-of-the-art photo upscaling). Output size scales with your input — a 1024×1024 input at 4× becomes a 4096×4096 output.
Increase image resolution by 2×, 4×, or 8× while keeping edges sharp and noise low. ImageSuite's upscaler runs the open-source Real-ESRGAN model (nightmareai/real-esrgan on Replicate), which is tuned for natural photos, screenshots, and illustrations. The result looks meaningfully sharper than a plain bicubic resize — particularly on small subjects, distant faces, and old web images.
AI Upscaler is a Pro feature ($9/mo flat). Free tools don't require signup; AI tools do.
JPG, PNG, or WebP. Very large inputs are downsized to fit the model's GPU memory window — that's normal.
2× for general use, 4× for printing small originals, 8× for restoring tiny thumbnails. Higher scales take longer.
Real-ESRGAN (nightmareai/real-esrgan), an open-source super-resolution GAN. ImageSuite calls it via Replicate. We name the model directly so you can compare against running it yourself.
Use 2× for general web sharpening. 4× when you want to print a phone-snap at 8×10 inches. 8× only for very small inputs — at high scales artifacts can creep in. Most users find 4× is the sweet spot.
Yes — Real-ESRGAN has a fixed GPU pixel budget per run. Inputs above ~2 megapixels are resized down before upscaling so the model can fit them. You still get a higher-resolution output than the original.
Real-ESRGAN is photo-tuned but handles clean illustrations well. For anime-specific super-resolution we may add Real-ESRGAN-Anime as a separate option later.