moeru-ai/airi · warning

[BG Removal Worker] WebGPU not available, falling back to WA

Error message

[BG Removal Worker] WebGPU not available, falling back to WASM

What it means

The background-removal worker mirrors the Whisper worker's device probe: when a request asks for device 'webgpu' (the default when request.device is unset) but detectWebGPUInWorker() finds no navigator.gpu, it logs this, downgrades to device 'wasm', and sets env.backends.onnx.wasm.proxy = false so the ONNX model runs directly on the worker thread. Background removal still works, just on CPU/WASM at lower throughput.

Source

Thrown at packages/stage-ui/src/workers/background-removal/worker.ts:145

        clearCancelled(requestId)
        return
      }
      const ready: ModelReadyResponse = {
        type: 'model-ready',
        requestId,
        modelId: MODEL_NAMES.BG_REMOVAL,
        device: resolvedDevice,
      }
      globalThis.postMessage(ready)
      return
    }

    // Auto-detect: if WebGPU was requested but unavailable, fall back to WASM
    let device = request.device ?? 'webgpu'
    if (device === 'webgpu') {
      const hasWebGPU = await detectWebGPUInWorker()
      if (!hasWebGPU) {
        console.warn('[BG Removal Worker] WebGPU not available, falling back to WASM')
        device = 'wasm'
      }
    }
    resolvedDevice = device as 'webgpu' | 'wasm' | 'cpu'

    env.backends.onnx.wasm!.proxy = false

    model = await AutoModel.from_pretrained(MODEL_ID, {
      device,
      progress_callback: (progress: any) => {
        sendProgress(requestId, progress?.progress ?? -1, progress?.status)
      },
    })

    processor = await AutoProcessor.from_pretrained(MODEL_ID, {})

    if (isCancelled(requestId)) {
      clearCancelled(requestId)

View on GitHub (pinned to 677329427f)

Solutions

1. Verify WebGPU availability in that browser (chrome://gpu or 'gpu' in navigator)
2. Update the browser / drivers if the hardware supports WebGPU
3. Pass device: 'wasm' in the request to skip the probe and the warning when GPU absence is expected
4. Use a smaller bg-removal model or lower-resolution input to offset WASM latency

Example fix

// before
worker.postMessage({ type: 'run', /* device defaults to 'webgpu' */ })

// after
const device = 'gpu' in navigator ? 'webgpu' : 'wasm'
worker.postMessage({ type: 'run', device })

const device: 'webgpu' | 'wasm' = 'gpu' in navigator ? 'webgpu' : 'wasm'
worker.postMessage({ type: 'run', requestId, device })

try {
  await runRemoval({ device: 'webgpu' })
}
catch {
  await runRemoval({ device: 'wasm' })
}

When it happens

Trigger: Any default request (or explicit device 'webgpu') in a browser/context without WebGPU: Firefox, older Chromium/Safari, driver blocklists, headless CI, VMs with software renderers, or worker scopes where navigator.gpu is not exposed.

Common situations: Users on WebGPU-less browsers experience slower background removal; automated screenshot pipelines; machines where the GPU driver is blocklisted.

Related errors

• [Whisper Worker] WebGPU not available, falling back to WASM
• [Whisper Worker] fp16 encoder failed, falling back to fp32:
• [Kokoro Worker] Failed with dtype=${attempt.dtype} device=${
• Failed to ensure MediaPipe WASM assets: ${wasmOutputDir} is
• Invalid VRAM override: ${bytes} (expected null or non-negati

AI-assisted analysis of moeru-ai/airi@677329427f (2026-08-18). Data as JSON: /api/errors/7a16c92983caf2cd. Report an issue: GitHub.