Upscale and denoise without uploading a pixel.
Detailed enlargement, computed locally.
A real AI model running on your device — no subscription, no API quota.
100% private
Your images never leave your device. All processing happens locally in your browser.
WebGPU acceleration
Inference uses your GPU when one is available, with automatic WASM fallback.
The Swin2SR model
Real-world super-resolution trained on BSRGAN degradation, so noisy photos enlarge instead of falling apart.
2× and 4× modes
Use 2× for speed or 4× for maximum detail. A denoise-only mode is included.
Two AI engines
Swin2SR rebuilds noisy real-world photos; a dedicated Real-ESRGAN model sharpens anime, illustrations, and line art (Pro). Each model downloads only the first time you pick it.
Batch, faces & formats
Queue up to 4 images per run on the free plan — 24 on Pro — and download everything as one ZIP. Export PNG, WebP, or JPEG with adjustable quality, with an optional GFPGAN face-restore pass for portraits (Pro).
Add, scale, download.
- 01
Upload an image
Drag and drop, paste, or browse. JPG, PNG, or WebP, up to your device's memory.
- 02
Pick a scale
Choose 2× or 4× super-resolution. The model runs locally through WebGPU.
- 03
Download the result
Download results one by one, or export the whole batch as a ZIP in PNG, WebP, or JPEG with adjustable quality. The models stay cached for the next run.
About the AI upscaler
A free AI image upscaler that enlarges photos by 2× or 4× while reconstructing detail rather than blurring it. Use it to rescue old low-resolution images, prepare small assets for print or high-DPI screens, or enlarge a downloaded picture that needs to fill a modern layout.
How it works
The tool runs Swin2SR, a state-of-the-art super-resolution model, in your browser through Transformers.js. On WebGPU-enabled browsers the model executes on your GPU; large images are processed in overlapping tiles so that even modest machines avoid running out of memory. The model hallucinates plausible sharp edges and texture where the source has none, which is why AI upscaling beats classic resampling on real-world photos.
Limits & requirements
First use downloads the model (a few tens of megabytes, cached after that). Upscaling is compute-intensive: a 4× pass on a large image may take a while on older laptops, and WebGPU is strongly recommended. Super-resolution adds plausible detail — it cannot recover information that was never captured, and text-heavy images may get minor artifacts.
Privacy
All inference happens locally. Your images are never transmitted, and nothing about your files is collected. Process personal photos, confidential designs, or client work with the same assurance.