LogoImage Upscaler Free
  • 功能
  • 博客
立即放大
LogoImage Upscaler Free

在浏览器本地运行的免费 AI 图片放大器

support@imageupscalerfree.com
产品
  • 功能
  • 放大到 4K
  • 动漫图片放大
  • 常见问题
公司
  • 关于我们
  • 联系我们
  • 博客
法律
  • Cookie政策
  • 开源致谢
  • 隐私政策
  • 服务条款
© 2026 Image Upscaler Free All Rights Reserved. · AI Toolz Dir

Open Source Credits

The open-source models and libraries that power ImageUpscalerFree, and their licences.

2026/09/11

Open Source Credits

Upscaling on ImageUpscalerFree runs entirely inside your browser. It is built on open-source work, credited here as those licences require.

Upscaling models

Both models are by Philip Hofmann (Phhofm/models) and are licensed under CC BY 4.0.

ModelUsed forSource
2xNomosUni_span_multijpg2× upscalingRelease page
4xNomosUni_span_multijpg4× upscalingRelease page

Modifications: the published ONNX files were altered in two ways. The output tensor's symbolic dimension names were renamed so they no longer collide with the input's, which otherwise prevents ONNX Runtime from allocating the larger output buffer. No weights were changed.

Both models use the SPAN architecture, from Swift Parameter-free Attention Network for Efficient Super-Resolution (Apache 2.0).

Runtime

ONNX Runtime Web by Microsoft, licensed under the MIT licence, executes the models in your browser using WebGPU where available and WebAssembly otherwise.

What this means for your images

Your image is never uploaded. The model files are downloaded to your device the first time you upscale something, and everything after that happens locally — which is also why the first run takes longer than later ones.