update notes and limit time

LHM/utils/ffmpeg_utils.py

@@ -46,13 +46,15 @@ def images_to_video(images, output_path, fps, gradio_codec: bool, verbose=False,
                 f"Frame shape mismatch: {frame.shape} vs {images.shape}"
             assert frame.min() >= 0 and frame.max() <= 255, \
                 f"Frame value out of range: {frame.min()} ~ {frame.max()}"
-            
         else:
             frame = images[i]
-        width, height = frame.shape[1], frame.shape[0]
+        # width, height = frame.shape[1], frame.shape[0]
         # reshape to limit the export time
-        if width > 1200 or height > 1200 or images.shape[0] > 400:
-            frames.append(cv2.resize(frame, (width // 2, height // 2)))
+        # if width > 1200 or height > 1200 or images.shape[0] > 400:
+        #     frames.append(cv2.resize(frame, (width // 2, height // 2)))
+        
+        # limit the frames directly @NOTE huggingface only！
+        frames = frames[:300]
     
     frames = np.stack(frames)
     iio.imwrite(output_path,frames,fps=fps,codec="libx264",pixelformat="yuv420p",bitrate=bitrate,macro_block_size=16)

app.py

@@ -644,7 +644,7 @@ def demo_lhm(pose_estimator, face_detector, lhm, cfg):
             """
         )
         gr.HTML(
-            """<p><h4 style="color: red;"> Notes: Please input full-body image in case of detection errors.</h4></p>"""
+            """<p><h4 style="color: red;"> Notes: Please input full-body image in case of detection errors. We simplify the pipeline in spaces: 1) using Rembg instead of SAM2; 2) limit the output video length to 10s; For best visual quality, try the inference code on Github instead.</h4></p>"""
         )
 
         # DISPLAY