Real-ESRGAN reconstructs detail as it enlarges, so a 2x or 4x result looks photographed, not stretched.
Classic interpolation — bicubic, Lanczos — has no idea what it is enlarging, so it averages neighbouring pixels and you get a bigger, softer version of the same photo. A super-resolution network was trained on millions of image pairs and predicts the detail that would have been there at the higher resolution: pore texture, fabric weave, the edge of a leaf.
ImageLabs uses Real-ESRGAN, the model most upscaling services are built on, and caps output at 4000 pixels on the long edge for stability. Large inputs are handled without the out-of-memory failures that plague browser-side upscalers.
Upload the image: JPG, PNG, WebP or HEIC. Small, soft or low-resolution sources benefit the most.
Choose 2x or 4K: 2x doubles the long edge; the 4K option targets a 4000-pixel output.
Download: Full resolution, no watermark, on every plan including the free tier.
Detail, not blur: Texture is reconstructed rather than smeared, so prints hold up at size.
Faces handled separately: Portraits can be routed through face restoration first so eyes and mouths stay sharp.
Predictable cost: 2 credits for 2x, 4 for 4K. Failed jobs are refunded automatically.
Can I upscale to 4K for free? Yes — 10 free credits per day covers multiple 2x upscales, and the 4K tier costs 4 credits.
Will it invent fake detail? Super-resolution always predicts. Real-ESRGAN is tuned to stay photographic, but a heavily degraded source will be interpreted rather than recovered.
What is the maximum output? 4000 pixels on the long edge, which covers 4K and most print sizes.