diff --git a/backend/src/plugins/llm_analysis/service.py b/backend/src/plugins/llm_analysis/service.py
index c2836604..5c4ca40a 100644
--- a/backend/src/plugins/llm_analysis/service.py
+++ b/backend/src/plugins/llm_analysis/service.py
@@ -1297,8 +1297,12 @@ class LLMClient:
             prompt = render_prompt(prompt_template, {"logs": log_text})
 
             # 2. Determine chunking
+            # Default to 8 images per chunk as a safe fallback when max_images is 0 or None
+            # (0 means probe failed — e.g. Kilo gateway doesn't support OpenAI image format)
+            DEFAULT_CHUNK_SIZE = 8
+            effective_max = max_images if (max_images is not None and max_images > 0) else DEFAULT_CHUNK_SIZE
             n_total = len(encoded_images)
-            chunk_size = max_images if (max_images and max_images > 0 and max_images < n_total) else n_total
+            chunk_size = effective_max if effective_max < n_total else n_total
             is_chunking = chunk_size < n_total
 
             if is_chunking: