Sadtalker Settings Tweaks

daras_ai_v2/lipsync_api.py

@@ -16,20 +16,44 @@ class LipsyncModel(Enum):
 
 class SadTalkerSettings(BaseModel):
     still: bool = Field(
-        False, title="Still (fewer head motion, works with preprocess 'full')"
+        True, title="Still (fewer head motion, works with preprocess 'full')"
     )
     preprocess: typing.Literal["crop", "extcrop", "resize", "full", "extfull"] = Field(
-        "crop", title="Preprocess"
+        "resize",
+        title="Preprocess",
+        description="SadTalker only generates 512x512 output. 'crop' handles this by cropping the input to 512x512. 'resize' scales down the input to fit 512x512 and scales it back up after lipsyncing (does not work well for full person images, better for portraits). 'full' processes the cropped region and pastes it back into the original input. 'extcrop' and 'extfull' are similar to 'crop' and 'full' but with extended cropping.",
+    )
+    pose_style: int = Field(
+        0,
+        title="Pose Style",
+        description="Random seed 0-45 inclusive that affects how the pose is animated.",
+    )
+    expression_scale: float = Field(
+        1.0,
+        title="Expression Scale",
+        description="Scale the amount of expression motion. 1.0 is normal, 0.5 is very reduced, and 2.0 is quite a lot.",
+    )
+    ref_eyeblink: FieldHttpUrl = Field(
+        None,
+        title="Reference Eyeblink",
+        description="Optional reference video for eyeblinks to make the eyebrow movement more natural.",
+    )
+    ref_pose: FieldHttpUrl = Field(
+        None,
+        title="Reference Pose",
+        description="Optional reference video to pose the head.",
     )
-    pose_style: int = Field(0, title="Pose Style")
-    expression_scale: float = Field(1.0, title="Expression Scale")
-    ref_eyeblink: FieldHttpUrl = Field(None, title="Reference Eyeblink")
-    ref_pose: FieldHttpUrl = Field(None, title="Reference Pose")
-    input_yaw: list[int] = Field(None, title="Input Yaw (comma separated)")
-    input_pitch: list[int] = Field(None, title="Input Pitch (comma separated)")
-    input_roll: list[int] = Field(None, title="Input Roll (comma separated)")
     # enhancer: typing.Literal["gfpgan", "RestoreFormer"] =None
     # background_enhancer: typing.Literal["realesrgan"] =None
+    input_yaw: list[int] = Field(
+        None, title="Input Yaw (comma separated)", deprecated=True
+    )
+    input_pitch: list[int] = Field(
+        None, title="Input Pitch (comma separated)", deprecated=True
+    )
+    input_roll: list[int] = Field(
+        None, title="Input Roll (comma separated)", deprecated=True
+    )
 
 
 class LipsyncSettings(BaseModel):

daras_ai_v2/lipsync_settings_widgets.py

@@ -39,9 +39,6 @@ def sadtalker_settings(settings: SadTalkerSettings):
         **field_label_val(settings, "expression_scale"),
     )
 
-    # st.selectbox("Face Enhancer", [None, "gfpgan", "RestoreFormer"], value=settings.enhancer)
-    # st.selectbox("Background Enhancer", [None, "realesrgan"], value=settings.background_enhancer)
-
     settings.ref_eyeblink = (
         st.file_uploader(
             **field_label_val(settings, "ref_eyeblink"),
@@ -51,45 +48,12 @@ def sadtalker_settings(settings: SadTalkerSettings):
     )
 
     settings.ref_pose = (
-        st.file_uploader("Reference Pose", value=settings.ref_pose, accept=[".mp4"])
-        or None
-    )
-
-    input_yaw = st.text_input(
-        "Input Yaw (comma separated)",
-        value=", ".join(map(str, settings.input_yaw or [])),
-    )
-    try:
-        settings.input_yaw = (
-            list(map(int, filter(None, input_yaw.strip().split(",")))) or None
-        )
-    except ValueError:
-        settings.input_yaw = None
-        st.error("Please enter comma separated integers for Input Yaw")
-
-    input_pitch = st.text_input(
-        "Input Pitch (comma separated)",
-        value=", ".join(map(str, settings.input_pitch or [])),
-    )
-    try:
-        settings.input_pitch = (
-            list(map(int, filter(None, input_pitch.strip().split(",")))) or None
+        st.file_uploader(
+            **field_label_val(settings, "ref_pose"),
+            accept=[".mp4"],
         )
-    except ValueError:
-        settings.input_pitch = None
-        st.error("Please enter comma separated integers for Input Pitch")
-
-    input_roll = st.text_input(
-        "Input Roll (comma separated)",
-        value=", ".join(map(str, settings.input_roll or [])),
+        or None
     )
-    try:
-        settings.input_roll = (
-            list(map(int, filter(None, input_roll.strip().split(",")))) or None
-        )
-    except ValueError:
-        settings.input_roll = None
-        st.error("Please enter comma separated integers for Input Roll")
 
 
 def wav2lip_settings():