diff --git a/modules/videoInferenceInputs.py b/modules/videoInferenceInputs.py
index 7072eb3..9cd271e 100644
--- a/modules/videoInferenceInputs.py
+++ b/modules/videoInferenceInputs.py
@@ -11,11 +11,11 @@ class videoInferenceInputs:
             "required": {},
             "optional": {
                 "Image": ("IMAGE", {
-                    "tooltip": "Portrait image for video generation. This will be the main subject that will speak and move in the generated video."
+                    "tooltip": "Portrait image for video generation (main subject that will speak and move). Mutually exclusive with Avatar: when Avatar is set, Image is ignored."
                 }),
                 "Avatar": ("STRING", {
                     "default": "",
-                    "tooltip": "HeyGen avatar ID for a photo avatar or video avatar. Mutually exclusive with image_url and image_asset_id.",
+                    "tooltip": "HeyGen avatar ID for a photo or video avatar. Mutually exclusive with Image: when set, Image is ignored; use one or the other.",
                 }),
                 "Frame Images": ("RUNWAREVIDEOINPUTSFRAMEIMAGES", {
                     "tooltip": "Connect the Runware Video Inputs Frame node to provide timeline-constrained frame images."
@@ -63,7 +63,10 @@ class videoInferenceInputs:
             }
         }
 
-    DESCRIPTION = "Configure custom inputs for Runware Video Inference with OmniHuman 1.5 support, including image, audio, video, and mask inputs."
+    DESCRIPTION = (
+        "Configure custom inputs for Runware Video Inference (image, audio, video, mask, avatar, etc.). "
+        "Avatar and Image are mutually exclusive: if Avatar is set, Image is ignored; provide only one."
+    )
     FUNCTION = "createInputs"
     RETURN_TYPES = ("RUNWAREVIDEOINFERENCEINPUTS",)
     RETURN_NAMES = ("Video Inference Inputs",)
@@ -89,10 +92,10 @@ class videoInferenceInputs:
 
         inputs = {}
 
-        if avatar is not None and avatar.strip() != "":
+        has_avatar = avatar is not None and avatar.strip() != ""
+        if has_avatar:
             inputs["avatar"] = avatar.strip()
-
-        if image is not None:
+        if image is not None and not has_avatar:
             inputs["image"] = rwUtils.convertTensor2IMG(image)
 
         # Handle audio: can be either a string (legacy), a single dict, or a list of audio inputs
diff --git a/modules/videoInferenceSpeechInput.py b/modules/videoInferenceSpeechInput.py
index 6895e49..c716cfa 100644
--- a/modules/videoInferenceSpeechInput.py
+++ b/modules/videoInferenceSpeechInput.py
@@ -69,18 +69,23 @@ class RunwareVideoInferenceSpeechInput:
     FUNCTION = "createSpeech"
     CATEGORY = "Runware"
     DESCRIPTION = (
-        "Configure speech synthesis for Runware Video Inference (voice and text). "
-        "Connect to Runware Video Inference speech input. Both voice and text must be set for speech to be included."
+        "Configure speech synthesis for Runware Video Inference. Connect to Runware Video Inference speech input. "
+        "Speech is only sent when both voice and text are enabled and non-empty; otherwise the node outputs nothing. "
+        "Optional fields (speed, pitch, language) are added only when their toggles are on."
     )
 
     def createSpeech(self, **kwargs) -> tuple:
-        """Build speech dict for genConfig[0]['speech']."""
-        
+        """Build speech dict for genConfig[0]['speech']. Only include non-empty values; return None unless both voice and text are present."""
         speech: Dict[str, Any] = {}
-        if kwargs.get("useVoice"):
-            speech["voice"] = kwargs.get("voice")
-        if kwargs.get("useText"):
-            speech["text"] = kwargs.get("text")
+
+        voice = (kwargs.get("voice") or "").strip()
+        if kwargs.get("useVoice") and voice:
+            speech["voice"] = voice
+
+        text = (kwargs.get("text") or "").strip()
+        if kwargs.get("useText") and text:
+            speech["text"] = text
+
         if kwargs.get("useSpeed"):
             speech["speed"] = float(kwargs.get("speed", 1.0))
         if kwargs.get("usePitch"):
@@ -90,7 +95,9 @@ class RunwareVideoInferenceSpeechInput:
             if lang:
                 speech["language"] = lang
 
-        return (speech or None,)
+        if "voice" not in speech or "text" not in speech:
+            return (None,)
+        return (speech,)
 
 
 NODE_CLASS_MAPPINGS = {