AI Image Upscaler 2×
Enhance a small picture with a trained on-device model and save a PNG at twice its dimensions.
1. Choose a small picture
PNG/JPEG/WebP · 5 MB · 262,144 pixels maximum (for example 512 × 512)
2. Enhance at twice the size
A trained model estimates missing detail locally. It can invent texture, blur text or change faces. Transparent areas become white; metadata is omitted. Keep the original and avoid using enhanced pictures as evidence.
3. Inspect and save
Your enhanced picture will appear here.
Practical example
A 256 × 128 picture becomes a 512 × 256 PNG. Inspect letters, faces and tile boundaries before sharing.
Limits · supported files
PNG/JPEG/WebP up to 5 MB, 262,144 input pixels and 1,024 pixels per side. WebGL-capable current browser required; three-minute processing budget. 64-pixel tiles with overlap use the MIT-licensed UpscalerJS default RDN 2× model. Inferred texture may be inaccurate; not recovery of authentic missing detail. Transparency becomes white and metadata is omitted. No guaranteed quality improvement or competitor parity.
Step-by-step guide
How to use AI Image Upscaler 2×Review the example, result checks and limitations before using your own inputs.