What is AI image upscaler suitable for?
This upscaler uses Transformers.js image-to-image pipelines with Swin2SR lightweight and classical models for general photo detail, plus APISR RRDB and GRL models tuned for anime and illustrations. Inference runs in a dedicated browser Worker, not on an application server.
WebGPU fp32 is attempted first for speed, with WASM q8 as the fallback. The first run downloads the selected ONNX model from Hugging Face; later runs can reuse it while the page remains open. Output is always a PNG, capped at 16 megapixels and 8192 pixels per side for safe browser memory use.
Supported formats: JPG, PNG, WebP; single file limit: 30 MB; output up to 16 MP and 8192 px per side.
How to use AI image upscaler
Step 1
Add an image
Choose or drop a JPG, PNG, or WebP file up to 30 MB.
Step 2
Pick a model
Use the fast 2× default, the detailed Swin2SR model, or an APISR anime model.
Step 3
Compare and save
Run the local upscale, compare the original and result, then download the PNG.
FAQ
Is my image uploaded?
No. Decoding and model inference happen in your browser Worker. The browser only downloads public model files from Hugging Face on first use.
Why is there an output limit?
Large AI outputs can exhaust browser memory. Results are limited to 16,000,000 pixels and 8192 pixels on either side, and the estimate is checked before processing starts.
What should I know about APISR?
APISR is a GPL-3.0 model with an upstream academic-use notice. Review both before using its output in a commercial workflow; the Swin2SR options do not carry that APISR notice.
Technology and data sources
- Transformers.js — Browser inference API and WebGPU/WASM runtime documentation.
- Swin2SR Lightweight 2× — Fast 2× ONNX model used as the default upscaler.
- Swin2SR Classical 2× — 2× Swin2SR model for detailed general-purpose reconstruction.
- APISR RRDB 2× — Anime-oriented 2× generator model with GPL-3.0 licensing.
- APISR GRL 4× — Anime-oriented 4× generator model with GPL-3.0 licensing.
- APISR project — Upstream model project, implementation details, and license context.
