diff --git a/tests/js/auto-formatter.test.js b/tests/js/auto-formatter.test.js
index 3281a5a..78acd61 100644
--- a/tests/js/auto-formatter.test.js
+++ b/tests/js/auto-formatter.test.js
@@ -41,6 +41,38 @@ describe('AutoFormatter Functions', () => {
             expect(shouldAutoFormat('X, 1.5, Y', mockNodeInfo('CLIPTextEncode', 'text'))).toBe(false);
         });
 
+        test('should return false for JSON-like content (avoid corrupting JSON commas)', () => {
+            const jsonText = `{
+  "quality_meta_year_safe": "masterpiece, best quality, score_9, year 2025, safe",
+  "count": "1girl",
+  "artist": "{{artist}}",
+  "tags": "tag1, tag2",
+  "width": 1024
+}`;
+
+            // JSON contains many "word + comma" patterns inside string values,
+            // but it must never be auto-formatted as prompt tags.
+            expect(shouldAutoFormat(jsonText, mockNodeInfo('SimpleChatTextInput', 'text'))).toBe(false);
+        });
+
+        test('should return true for prompts containing comfyui-prompt-control scheduling syntax', () => {
+            // Range expression: [before:during:after:start,end]
+            expect(shouldAutoFormat('[black::white:0.3,0.7], 1girl, blue hair', mockNodeInfo('CLIPTextEncode', 'text'))).toBe(true);
+
+            // Alternating: [a|b:pct]
+            expect(shouldAutoFormat('[cat|dog:0.1], 1girl, solo', mockNodeInfo('CLIPTextEncode', 'text'))).toBe(true);
+
+            // Sequence: [SEQ:a:N1:b:N2:...]
+            expect(shouldAutoFormat('[SEQ:red:0.3:blue:0.6:green:0.9], 1girl', mockNodeInfo('CLIPTextEncode', 'text'))).toBe(true);
+
+            // LoRA scheduling
+            expect(shouldAutoFormat('<lora:fulllora:1> [<lora:partiallora:1>::0.5], 1girl', mockNodeInfo('CLIPTextEncode', 'text'))).toBe(true);
+
+            // Nested scheduling
+            expect(shouldAutoFormat('[red:[blue::0.7]:0.5] cat, 1girl', mockNodeInfo('CLIPTextEncode', 'text'))).toBe(true);
+        });
+
+
         test('should return true for text with "word + comma" pattern', () => {
             expect(shouldAutoFormat('1girl, blue hair,', mockNodeInfo('CLIPTextEncode', 'text'))).toBe(true);
             expect(shouldAutoFormat('tag1, tag2', mockNodeInfo('CLIPTextEncode', 'text'))).toBe(true);
diff --git a/web/js/auto-formatter.js b/web/js/auto-formatter.js
index 9c586ae..8a76993 100644
--- a/web/js/auto-formatter.js
+++ b/web/js/auto-formatter.js
@@ -1,6 +1,26 @@
 import { settingValues } from './settings.js';
 import { NodeInfo } from './node-info.js';
 
+/**
+ * JSON content detection with actual parsing.
+ * 
+ * @param {string} text - The text to check for JSON content.
+ * @return {boolean} - True if the text is valid JSON, false otherwise.
+ */
+function isJsonContent(text) {
+    if (!text) return false;
+    const t = text.trim();
+    if (t.length < 2) return false;
+    if (t[0] !== '{' && t[0] !== '[') return false;
+
+    try {
+        JSON.parse(t);
+        return true;
+    } catch {
+        return false;
+    }
+}
+
 
 /**
  * Determines if the text content should be auto-formatted.
@@ -8,10 +28,11 @@ import { NodeInfo } from './node-info.js';
  * Format conditions:
  * 1. Skip formatting if node is in blocklist
  * 2. If text is empty after trimming, format only if trimSurroundingSpaces is enabled, otherwise don't format
- * 3. Skip formatting if text contains only numbers or single letters (separated by commas)
- * 4. Format if text contains "word + comma" pattern
- * 5. Format if trimSurroundingSpaces is enabled and there are surrounding spaces or line breaks
- * 6. Otherwise, don't format
+ * 3. Skip formatting if text is valid JSON
+ * 4. Skip formatting if text contains only numbers or single letters (separated by commas)
+ * 5. Format if text contains "word + comma" pattern
+ * 6. Format if trimSurroundingSpaces is enabled and there are surrounding spaces or line breaks
+ * 7. Otherwise, don't format
  *
  * @param {NodeInfo} nodeInfo - The node information.
  * @returns {boolean} - True if the text should be formatted, false otherwise.
@@ -43,7 +64,12 @@ function shouldAutoFormat(text, nodeInfo) {
         return settingValues.trimSurroundingSpaces;
     }
 
-    // 3. Check if the text is purely numeric data or single-letter placeholders with commas
+    // 3. Skip JSON content (do not destroy JSON commas)
+    if (isJsonContent(trimmedText)) {
+        return false;
+    }
+
+    // 4. Check if the text is purely numeric data or single-letter placeholders with commas
     // (e.g., "0,0,0,1,1,1" or "0.5, -1.2, 0.8" or "A,B,R" for LoRA Block Weight)
     const elements = trimmedText.split(',').map(el => el.trim());
     const isSingleLetterOrNumeric = elements.every(el => {
@@ -56,13 +82,13 @@ function shouldAutoFormat(text, nodeInfo) {
         return false; // Don't format numeric data or single-letter template patterns
     }
 
-    // 4. If text contains "word + comma" pattern, format it
-    const wordCommaPattern = /\w+\s*,/g;
+    // 5. If text contains "word + comma" pattern, format it
+    const wordCommaPattern = /[^\s,]+\s*,/g;
     if (trimmedText.match(wordCommaPattern)) {
         return true;
     }
 
-    // 5. If trimSurroundingSpaces is enabled and there are surrounding spaces, format to trim them
+    // 6. If trimSurroundingSpaces is enabled and there are surrounding spaces, format to trim them
     if (settingValues.trimSurroundingSpaces && text !== trimmedText) {
         return true;
     }