Tinier upscale models

ONNX exports of the official Real-ESRGAN weights, used by the in-browser AI upscaler at tinier.app/upscale-image (onnxruntime-web, WebGPU/WASM).

File	Source weights	Params	Size
`realesr-general-x4v3.onnx`	realesr-general-x4v3.pth (SRVGGNetCompact)	1.2M	4.6 MB
`realesrgan-x4plus.onnx`	RealESRGAN_x4plus.pth (RRDBNet)	16.7M	64 MB

Export details: fp32, opset 17, dynamic H/W axes. Contract: input float32 [1, 3, h, w], RGB in [0, 1], NCHW → output float32 [1, 3, 4h, 4w].

Both exports are numerically verified against the PyTorch originals (max abs diff ~1e-6). Conversion script: scripts/export-onnx-models.py in the Tinier repo.

License: BSD-3-Clause, inherited from the upstream Real-ESRGAN weights.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support