AI upscaling uses neural networks (trained on millions of images) to intelligently reconstruct high-resolution pixels when enlarging an image. Unlike traditional methods like bicubic interpolation — which blur the image — AI models recreate realistic textures like hair, fabric, skin, and bark that weren't visible in the lower-resolution original.
Real-ESRGAN is a CNN-GAN model that's fast (15–30 seconds) and very clean. It works well for photos, screenshots, anime, and illustrations. Clarity Upscaler is Stable Diffusion-based and takes longer (45–90 seconds) but adds genuinely generated detail and texture — better for portraits, nature photography, and artistic images where maximum quality matters.
2× roughly doubles linear resolution (e.g., 1000px → 2000px). 4× quadruples it (1000px → 4000px). 8× is for extreme enlargement — good for very small source images. As a rule: if your image looks decent at normal size but is too small for your use case, 2× or 4× is usually enough. 8× is for stamps, thumbnails, or very degraded sources.
Yes. You get 10 free upscales per day with no account required. A Pro plan with unlimited access is in development.
JPEG, PNG, and WebP. Maximum file size is 10 MB per upload.
Replicate processes the image at the selected scale factor — so a 2000×1500px image at 4× becomes 8000×6000px. There is no hard cap on output resolution, but very large inputs may time out. We recommend keeping input images under 4000px on the longest side.
Real-ESRGAN: typically 15–30 seconds. Clarity Upscaler: typically 45–90 seconds. Processing time depends on server load and image size.
Yes. The AI-upscaled output of an image you own is yours. Upscaler.online claims no ownership. Always ensure you have rights to the original image you upload.
Images are transmitted to Replicate for processing. Replicate stores them temporarily while the prediction runs. We do not store images on Upscaler.online servers. See our Privacy Policy for full details.
When you enable face enhancement with Real-ESRGAN, the GFPGAN face restoration model runs alongside the upscaling — detecting faces and applying additional sharpening, texture restoration, and correction to facial features. Highly recommended for old or degraded portrait photos.
Creativity controls how aggressively the AI adds new detail: 0.3 = subtle enhancements, 0.9 = the AI substantially regenerates texture. Resemblance controls how closely the output matches the original composition: 0.3 = creative reinterpretation, 1.6 = very faithful to the original. We recommend starting at creativity: 0.35, resemblance: 0.6 for most images.
Still have questions?
Contact Support →