How AI Image Upscaler Works
2× and 4× super-resolution with automatic source routing — the right model for clean photos, compressed photos, and flat art each, because one model cannot serve all three.
Try it free →Step-by-step process
- 1
Source analysis and routing
The upscaler inspects your image first. Clean photos go to Swin2SR (faithful detail synthesis); compressed or previously-downscaled photos go to a real-world restoration model trained on actual degradation, which is what recovers sharpness instead of enlarging mush; logos and flat art skip photo-AI entirely for an edge-preserving scaler, keeping fills flat and lines crisp. A Source selector overrides the automatic choice.
- 2
Tile-based processing
AI passes divide the image into overlapping tiles. Each tile runs through the neural network independently and the results are assembled with seamless blending at tile boundaries to avoid visible seams.
- 3
Honest limits, full-resolution export
Requests beyond 8× of the source are capped with a notice rather than silently delivering blur — a 64px icon does not contain 4096px of photographic detail. The result exports as PNG at up to 4096 px per side (4×) or 8192 px (2×).
Frequently asked questions
What is the difference between AI mode and Instant mode?
AI mode uses a neural network (Swin2SR) to synthesise new pixels — it produces sharper, more detailed results especially for faces, text, and fine textures. Instant mode uses Lanczos mathematical upscaling plus an unsharp mask — it is much faster but does not add new detail.
How long does 4× upscaling take?
A 1000 × 1000 px image at 4× takes approximately 30–90 seconds in AI mode on a modern GPU via WebGPU. Larger images take longer proportionally due to the tile count.
Is transparency preserved?
Yes. The upscaler processes the RGB channels and applies the same upscale to the alpha channel separately, preserving transparency in PNG output.
Why does the output look identical to the original?
This can happen if your image was already high resolution — upscaling a sharp 2000 px image to 4000 px adds size but not visible detail. AI upscaling is most impactful on low-resolution or compressed source images.