--- before/wechat_rpa/model_protocol.py
+++ wechat_rpa/model_protocol.py
@@ -51,6 +51,24 @@
 
 
 ENDPOINT_MODES = ("auto", "exact")
+
+
+def chat_config_error(kind: str, base_url: str, model: str = "") -> str:
+    """Identify known video configurations before sending a chat request."""
+    if detect_kind(kind, base_url) != "openai":
+        return ""
+    path = (urlparse(base_url or "").path or "").lower().rstrip("/")
+    model_name = str(model or "").strip().lower().rsplit("/", 1)[-1]
+    video_path = path.endswith("/contents/generations/tasks") or "/contents/generations/tasks/" in path
+    video_model = model_name.startswith(("doubao-seedance-", "seedance-"))
+    if video_path or video_model:
+        return (
+            "配置的是 Seedance 视频生成模型或视频任务接口，不能用于客服聊天。"
+            "请在火山方舟开通支持 Chat API 的文本/视觉理解模型，填写其模型 ID 或推理接入点 ID；"
+            "接口地址使用 https://ark.cn-beijing.volces.com/api/v3/chat/completions。"
+            "完整地址开关只控制路径拼接，不会转换接口协议。"
+        )
+    return ""
 
 
 def endpoint_url(kind: str, base_url: str, mode: str = "auto") -> str:
@@ -196,6 +214,9 @@
 ) -> dict:
     """按各家的形状拼请求体。"""
     kind = detect_kind(kind, base_url)
+    config_error = chat_config_error(kind, base_url, model)
+    if config_error:
+        raise ValueError(config_error)
     if kind == "claude":
         system_text, turns = claude_messages(messages)
         payload = {