diff --git a/js/msr_character.js b/js/msr_character.js
index 15069e7..7663159 100644
--- a/js/msr_character.js
+++ b/js/msr_character.js
@@ -60,7 +60,22 @@ function resolveCandidateImageUrls(originNode) {
 async function imageInputToDataUrl(node, inputName) {
   const originNode = getOriginNodeForInput(node, inputName);
   const candidateUrls = resolveCandidateImageUrls(originNode);
-  if (!candidateUrls.length) return null;
+  if (!candidateUrls.length) {
+    return {
+      dataUrl: null,
+      debug: {
+        inputName,
+        originNodeId: originNode?.id ?? null,
+        originNodeType: originNode?.type ?? null,
+        candidateUrls: [],
+        selectedUrl: null,
+        mimeType: null,
+        blobBytes: 0,
+        dataUrlLength: 0,
+        error: "No candidate image URLs found",
+      },
+    };
+  }
 
   for (const imageUrl of candidateUrls) {
     try {
@@ -72,17 +87,44 @@ async function imageInputToDataUrl(node, inputName) {
       if (!blob.type.startsWith("image/")) {
         throw new Error(`Unexpected blob type: ${blob.type || "unknown"}`);
       }
-      return await new Promise((resolve) => {
+      const dataUrl = await new Promise((resolve) => {
         const reader = new FileReader();
         reader.onloadend = () => resolve(reader.result);
         reader.readAsDataURL(blob);
       });
+      return {
+        dataUrl,
+        debug: {
+          inputName,
+          originNodeId: originNode?.id ?? null,
+          originNodeType: originNode?.type ?? null,
+          candidateUrls,
+          selectedUrl: imageUrl,
+          mimeType: blob.type || null,
+          blobBytes: blob.size || 0,
+          dataUrlLength: typeof dataUrl === "string" ? dataUrl.length : 0,
+          error: null,
+        },
+      };
     } catch (error) {
       console.warn("[MSRCharacter] Failed candidate image source", imageUrl, error);
     }
   }
 
-  return null;
+  return {
+    dataUrl: null,
+    debug: {
+      inputName,
+      originNodeId: originNode?.id ?? null,
+      originNodeType: originNode?.type ?? null,
+      candidateUrls,
+      selectedUrl: null,
+      mimeType: null,
+      blobBytes: 0,
+      dataUrlLength: 0,
+      error: "All candidate image URLs failed",
+    },
+  };
 }
 
 function setWidgetValue(node, widget, value) {
@@ -330,11 +372,13 @@ async function analyzeCharacterNode(node, buttonWidget) {
   node.setDirtyCanvas?.(true, true);
 
   try {
+    const imageResults = await Promise.all([
+      imageInputToDataUrl(node, "image_1"),
+      imageInputToDataUrl(node, "image_2"),
+    ]);
+    const imageDebug = imageResults.map((result) => result?.debug || null).filter(Boolean);
     const images = (
-      await Promise.all([
-        imageInputToDataUrl(node, "image_1"),
-        imageInputToDataUrl(node, "image_2"),
-      ])
+      imageResults.map((result) => result?.dataUrl || null)
     ).filter(Boolean);
 
     if (!images.length) {
@@ -348,10 +392,15 @@ async function analyzeCharacterNode(node, buttonWidget) {
         base_url: baseUrlWidget?.value || "",
         model: modelWidget?.value || "",
         image_b64: images,
+        image_debug: imageDebug,
       }),
     });
     const result = await response.json();
     if (result.status !== "success") {
+      console.warn("[MSRCharacter] analyze debug", {
+        sentImages: imageDebug,
+        response: result,
+      });
       throw new Error(result.message || "Unknown analysis error");
     }
 
diff --git a/ltx_director.py b/ltx_director.py
index f9fb690..acf1449 100644
--- a/ltx_director.py
+++ b/ltx_director.py
@@ -549,10 +549,11 @@ def _resolve_provider(data):
 async def analyze_character_endpoint(request):
     try:
         import aiohttp
-        data = await request.json()
-        image_b64 = data.get("image_b64", "")
-        char_index = int(data.get("char_index", 0))
-        provider, base_url, model_name = _resolve_provider(data)
+        data = await request.json()
+        image_b64 = data.get("image_b64", "")
+        image_debug = data.get("image_debug") or []
+        char_index = int(data.get("char_index", 0))
+        provider, base_url, model_name = _resolve_provider(data)
 
         if provider == "off":
             return web.json_response({"status": "error", "message": "Analyze is set to Off / Manual."})
@@ -639,6 +640,8 @@ async def analyze_character_endpoint(request):
                             "status": "error",
                             "message": "Ollama returned only a truncated analysis response.",
                             "debug_candidates": debug_candidates,
+                            "image_debug": image_debug,
+                            "image_lengths": [len(b64) for b64 in cleaned_b64_list],
                             "description": generated_text,
                         })
                 else: