Self-hosted Swin2SR model
Swin2SR Lightweight 2x
A featherweight SwinV2 model that doubles resolution quickly and keeps the full fp32 weights for maximum fidelity at this size.
Model specifications
- Scale factor
- 2x
- Precision
- fp32
- Parameters
- 1.01M · fp32
- Exact size
- 8,078,888 bytes
- Best for
- Speed on any device
- License
- Apache-2.0
SHA-256
c2abbfe0cc8e685b5e11964970f8ebe3d24072e904fd5545a30ac31ec1e110db Model architecture and pretrained weights are attributed to caidas/swin2SR-lightweight-x2-64. The ONNX conversion is hosted at Xenova/swin2SR-lightweight-x2-64. Distributed under the Apache-2.0 license.
Use the self-hosted file
This is the same same-origin asset requested by WKO AI's studio. Remote model fallback is disabled, so a missing mirror fails explicitly.
import { env, pipeline } from '@huggingface/transformers';
env.allowLocalModels = true;
env.allowRemoteModels = false;
env.localModelPath = '/models/';
const upscale = await pipeline(
'image-to-image',
'Xenova/swin2SR-lightweight-x2-64',
{ dtype: 'fp32' }
);
const output = await upscale(image);