@heripo/pdf-parser 0.1.12 → 0.1.13

package/dist/index.js

@@ -1756,7 +1756,8 @@ var VlmTextCorrector = class {
               },
               {
                 type: "image",
-                image: `data:image/png;base64,${imageBase64}`
+                image: imageBase64,
+                mediaType: "image/png"
               }
             ]
           }
@@ -1968,7 +1969,7 @@ var VlmTextCorrector = class {
    */
   readPageImage(outputDir, pageNo) {
     const imagePath = join4(outputDir, "pages", `page_${pageNo - 1}.png`);
-    return readFileSync(imagePath).toString("base64");
+    return new Uint8Array(readFileSync(imagePath));
   }
   /**
    * Apply VLM corrections to the DoclingDocument.
@@ -2237,7 +2238,7 @@ var OcrStrategySampler = class {
     this.logger.debug(
       `[OcrStrategySampler] Analyzing page ${pageNo} for Korean-Hanja mix and language...`
     );
-    const base64Image = readFileSync2(pageFile).toString("base64");
+    const imageData = new Uint8Array(readFileSync2(pageFile));
     const messages = [
       {
         role: "user",
@@ -2245,7 +2246,8 @@ var OcrStrategySampler = class {
           { type: "text", text: KOREAN_HANJA_MIX_PROMPT },
           {
             type: "image",
-            image: `data:image/png;base64,${base64Image}`
+            image: imageData,
+            mediaType: "image/png"
           }
         ]
       }