diff --git a/model_specs/dramabox.json b/model_specs/dramabox.json
index ed494c8d3..3ecbe701b 100644
--- a/model_specs/dramabox.json
+++ b/model_specs/dramabox.json
@@ -57,7 +57,7 @@
       {
         "name": "num_inference_steps",
         "type": "int",
-        "description": "Diffusion sampling step count; default comes from config.json, 30 in the current package.",
+        "description": "Diffusion sampling step count; default 30.",
         "required": false,
         "min": 1,
         "default": 30
@@ -65,7 +65,7 @@
       {
         "name": "guidance_scale",
         "type": "float",
-        "description": "Classifier-free guidance scale; default comes from config.json, 2.5 in the current package. Values greater than 1 enable CFG.",
+        "description": "Classifier-free guidance scale; default 2.5. Values greater than 1 enable CFG.",
         "required": false,
         "min": 0.0,
         "default": 2.5
@@ -73,7 +73,7 @@
       {
         "name": "spatio_temporal_guidance_scale",
         "type": "float",
-        "description": "Spatio-temporal guidance scale; default comes from config.json, 1.5 in the current package. Values greater than 0 enable STG.",
+        "description": "Spatio-temporal guidance scale; default 1.5. Values greater than 0 enable STG.",
         "required": false,
         "min": 0.0,
         "default": 1.5
@@ -81,7 +81,7 @@
       {
         "name": "duration_scale",
         "type": "float",
-        "description": "Multiplier applied to the estimated prompt duration when duration_sec is 0; default comes from config.json, 1.1 in the current package.",
+        "description": "Multiplier applied to the estimated prompt duration when duration_sec is 0; default 1.1.",
         "required": false,
         "min": 0.0,
         "default": 1.1
@@ -89,7 +89,7 @@
       {
         "name": "reference_duration_sec",
         "type": "float",
-        "description": "Reference voice crop/repeat duration in seconds; default comes from config.json, 10.0 in the current package.",
+        "description": "Reference voice crop/repeat duration in seconds; default 10.0.",
         "required": false,
         "min": 0.0,
         "default": 10.0
diff --git a/model_specs/fish_audio.json b/model_specs/fish_audio.json
index 9a15337c7..462bb4320 100644
--- a/model_specs/fish_audio.json
+++ b/model_specs/fish_audio.json
@@ -29,7 +29,7 @@
       {
         "name": "reference_text",
         "type": "string",
-        "description": "Reference transcript used with speaker reference audio.",
+        "description": "Reference transcript used with speaker reference audio. Required whenever the request carries inline reference audio; the model rejects the request without it.",
         "required": false
       },
       {
diff --git a/model_specs/outetts.json b/model_specs/outetts.json
index fc8e7dedc..d8f7d44ea 100644
--- a/model_specs/outetts.json
+++ b/model_specs/outetts.json
@@ -107,7 +107,7 @@
       {
         "name": "reference_text",
         "type": "string",
-        "description": "Transcript matching the reference voice audio for voice cloning.",
+        "description": "Transcript matching the reference voice audio for voice cloning. Required whenever reference audio is supplied; the model rejects voice cloning without it.",
         "required": false
       },
       {