diff --git a/ltx_director.py b/ltx_director.py
index e06098a..3fb342f 100644
--- a/ltx_director.py
+++ b/ltx_director.py
@@ -439,9 +439,10 @@ _PROVIDER_DEFAULTS = {
 }
 
 _ANALYZE_PROMPT = (
-    "Describe the character's physical appearance in two concise sentences. "
-    "Specify their hair color/style, face details, and their clothing type/color. "
-    "Keep the entire response very brief."
+    "Describe only the visible character in exactly two short sentences. "
+    "Mention hair, face, standout physical traits, and clothing. "
+    "Output only the description itself. Do not explain, do not list requirements, "
+    "do not say what the user wants, and do not include headings, bullets, or notes."
 )
 
 
@@ -512,58 +513,6 @@ def _analysis_text_is_usable(text: str) -> bool:
     return True
 
 
-def _sanitize_analysis_text(text: str) -> str:
-    text = (text or "").strip()
-    if not text:
-        return ""
-
-    if "<think>" in text:
-        text = text.split("</think>")[-1].strip()
-
-    cleaned_lines = []
-    for raw_line in text.splitlines():
-        line = raw_line.strip().replace("**", "")
-        if not line:
-            continue
-        lower = line.lower()
-        if lower.startswith("the user wants"):
-            continue
-        if lower.startswith("key elements to include"):
-            continue
-        if lower.startswith("drafting the sentences"):
-            continue
-        if lower.startswith("sentence 1"):
-            continue
-        if lower.startswith("sentence 2"):
-            continue
-        if line.startswith("*"):
-            continue
-        if re.match(r"^\d+\.\s", line):
-            continue
-        cleaned_lines.append(line)
-
-    text = " ".join(cleaned_lines).strip()
-    text = re.sub(r"\s+", " ", text)
-    if not text:
-        return ""
-
-    sentences = re.split(r"(?<=[.!?])\s+", text)
-    descriptive = []
-    for sentence in sentences:
-        sentence = sentence.strip()
-        if not sentence:
-            continue
-        lower = sentence.lower()
-        if "the user wants" in lower or "key elements" in lower or "drafting the sentences" in lower:
-            continue
-        descriptive.append(sentence)
-
-    if descriptive:
-        return " ".join(descriptive[-2:]).strip()
-
-    return text
-
-
 def _compress_analysis_image_b64(b64_payload: str, max_dim: int = 768, quality: int = 82) -> str:
     """Shrink analysis images so multimodal providers do not burn their full context on pixels."""
     try:
@@ -726,7 +675,8 @@ async def analyze_character_endpoint(request):
                 "message": f"Could not connect to {provider} at {base_url}. Make sure the server is running and reachable.",
             })
 
-        generated_text = _sanitize_analysis_text(generated_text)
+        if "<think>" in generated_text:
+            generated_text = generated_text.split("</think>")[-1].strip()
 
         log.info("[LTXDirector] Analysis complete: %s", generated_text)
         return web.json_response({"status": "success", "description": generated_text})