Upscale anime art with Hugging Face Xenova models.
Source
Original: -Dimensions: -
First run downloads the model from Hugging Face.
Run
Load an image to begin.
Scale 2xFull imageNo output yetNo file yet
Original
Upscaled
Model settings
2x RRDB GANRuntime
WebGPU not available in this browser. WASM uses multi-threaded CPU when available.
Data type
Q4 is the fastest overall. Q4F16 targets WebGPU, FP32 keeps the best detail, and UINT8 is fastest on CPU.
Memory
Auto release keeps memory low by unloading weights after each run.
Tiled upscale
Tiled mode processes the image in smaller chunks to reduce memory.
Upscale model
Runtime diagnostics
Cross-origin isolated-
SharedArrayBuffer-
WebGPU-
WebGPU F16 support-
WebGPU F16 enabled-
Hardware threads-
WASM threads-
WASM multi-threading requires cross-origin isolation (COOP/COEP).
Model source (Hugging Face)