Skip to content

The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.

[Partner Nodes] fix: respect VideoSlice trim when resizing videos (#14213)

Alexander Piskun0b610bd63a3e
diff --git a/comfy_api/latest/_input/video_types.py b/comfy_api/latest/_input/video_types.py
index 451e9526e..8fff52c16 100644
--- a/comfy_api/latest/_input/video_types.py
+++ b/comfy_api/latest/_input/video_types.py
@@ -65,6 +65,12 @@ class VideoInput(ABC):
buffer.seek(0)
return buffer
+ def get_active_trim_window(self) -> tuple[float, float]:
+ """Return the active trim as ``(start_time, duration)`` in seconds (start_time normalized
+ to ``>= 0``; ``duration == 0`` means "until the end"). Default: no trim; trimmable subclasses override.
+ """
+ return 0.0, 0.0
+
# Provide a default implementation, but subclasses can provide optimized versions
# if possible.
def get_dimensions(self) -> tuple[int, int]:
diff --git a/comfy_api/latest/_input_impl/video_types.py b/comfy_api/latest/_input_impl/video_types.py
index 99e67d363..4a12ff9c1 100644
--- a/comfy_api/latest/_input_impl/video_types.py
+++ b/comfy_api/latest/_input_impl/video_types.py
@@ -75,6 +75,12 @@ class VideoFromFile(VideoInput):
self.__file.seek(0)
return self.__file
+ def get_active_trim_window(self) -> tuple[float, float]:
+ start_time = self.__start_time
+ if start_time < 0:
+ start_time = max(self._get_raw_duration() + start_time, 0.0)
+ return float(start_time), float(self.__duration)
+
def get_dimensions(self) -> tuple[int, int]:
"""
Returns the dimensions of the video input.
diff --git a/comfy_api_nodes/util/conversions.py b/comfy_api_nodes/util/conversions.py
index 5738df57f..a1b5d599c 100644
--- a/comfy_api_nodes/util/conversions.py
+++ b/comfy_api_nodes/util/conversions.py
@@ -469,6 +469,11 @@ def _apply_video_scale(video: Input.Video, scale_dims: tuple[int, int]) -> Input
input_container = None
output_container = None
+ # get_stream_source() is untrimmed, so apply the trim window in this same pass.
+ # start_time is normalized (>= 0); duration == 0 means "until the end".
+ start_time, duration = video.get_active_trim_window()
+ trimming = bool(start_time or duration)
+
try:
input_source = video.get_stream_source()
input_container = av.open(input_source, mode="r")
@@ -487,16 +492,45 @@ def _apply_video_scale(video: Input.Video, scale_dims: tuple[int, int]) -> Input
audio_stream.layout = stream.layout
break
+ in_video = input_container.streams.video[0]
+ start_pts = int(start_time / in_video.time_base) if trimming else 0
+ end_pts = int((start_time + duration) / in_video.time_base) if duration else None
+ if start_pts:
+ input_container.seek(start_pts, stream=in_video)
+
+ encoded = 0
for frame in input_container.decode(video=0):
+ if trimming:
+ if frame.pts is None or frame.pts < start_pts:
+ continue
+ if end_pts is not None and frame.pts >= end_pts:
+ break
frame = frame.reformat(width=out_w, height=out_h, format="yuv420p")
+ # Re-wrap as a fresh frame: dropping irregular source timestamps (VFR/AVI/GIF/...)
+ # lets the encoder assign clean ones and avoids mp4 muxer errors.
+ frame = av.VideoFrame.from_ndarray(frame.to_ndarray(format="yuv420p"), format="yuv420p")
for packet in video_stream.encode(frame):
output_container.mux(packet)
+ encoded += 1
for packet in video_stream.encode():
output_container.mux(packet)
+ if encoded == 0:
+ raise ValueError(
+ f"resize produced no frames (start_time={start_time}, duration={duration} "
+ "selected nothing from the source)"
+ )
+
if audio_stream is not None:
input_container.seek(0)
for audio_frame in input_container.decode(audio=0):
+ if trimming:
+ if audio_frame.time is None or audio_frame.time < start_time:
+ continue
+ if duration and audio_frame.time > start_time + duration:
+ break
+ # Carry odd audio time bases the mp4 muxer rejects; reset pts, encoder assigns clean ones (MP3-in-AVI)
+ audio_frame.pts = None
for packet in audio_stream.encode(audio_frame):
output_container.mux(packet)
for packet in audio_stream.encode():