mozilla/pdf.js · error · Error

Error from the AI service.

Error message

Error from the AI service.

What it means

Thrown by StampEditor.mlGuessAltText when the 'altText' guess response carries a truthy `error` field. Unlike the no-response case, the ML service did reply, but the payload signals a backend-side failure (model error, bad input, rate limit, etc.). The library surfaces this as a distinct, non-recoverable error for that guess attempt.

Source

Thrown at src/display/editor/stamp.js:186

      throw new Error("ML isn't enabled for alt text.");
    }
    const { data, width, height } =
      imageData ||
      this.copyCanvas(null, null, /* createImageData = */ true).imageData;
    const response = await mlManager.guess({
      name: "altText",
      request: {
        data,
        width,
        height,
        channels: data.length / (width * height),
      },
    });
    if (!response) {
      throw new Error("No response from the AI service.");
    }
    if (response.error) {
      throw new Error("Error from the AI service.");
    }
    if (response.cancel) {
      return null;
    }
    if (!response.output) {
      throw new Error("No valid response from the AI service.");
    }
    const altText = response.output;
    await this.setGuessedAltText(altText);
    if (updateAltTextData && !this.hasAltTextData()) {
      this.altTextData = { alt: altText, decorative: false };
    }
    return altText;
  }

  #getBitmap() {
    if (this.#bitmapId) {
      this._uiManager.enableWaiting(true);

View on GitHub (pinned to 5903d58d58)

Solutions

1. Inspect response.error from your mlManager to identify the backend-reported cause.
2. Verify the imageData passed in has consistent width, height, and channels (data.length === width*height*channels).
3. Retry the guess after a short delay if the backend error is transient.
4. Catch the error and fall back to the manual alt-text dialog so the editor stays usable.

Example fix

// before
const alt = await editor.mlGuessAltText();
// after
try {
  const alt = await editor.mlGuessAltText();
} catch (e) {
  if (e.message === 'Error from the AI service.') {
    // log backend cause and fall back
  }
}

if (imageData.width <= 0 || imageData.height <= 0 || imageData.data.length === 0) {
  // do not call mlGuessAltText with bad image data
}

function isValidImageDataForGuess({ data, width, height }) {
  return width > 0 && height > 0 && data.length === width * height * (data.length / (width * height));
}

try {
  const alt = await editor.mlGuessAltText();
} catch (e) {
  if (e.message.startsWith('Error from the AI service.')) {
    // log and fall back to manual entry
  }
}

When it happens

Trigger: mlManager.guess({ name:'altText', request:{data,width,height,channels} }) resolves to an object whose `error` property is truthy (e.g. { error: 'model timeout' }).

Common situations: The image data payload is malformed or too large; the deployed alt-text model rejects the input; the backend is rate-limiting or temporarily failing; channels (data.length/(width*height)) computes to an unexpected value due to a zero-size canvas.

Related errors

• No response from the AI service.
• No valid response from the AI service.
• No ML.
• ML isn't enabled for alt text.

AI-assisted analysis of mozilla/pdf.js@5903d58d58 (2026-08-13). Data as JSON: /api/errors/b654a80ab473b866. Report an issue: GitHub.