import gradio as gr import argparse import os from musepose_inference import MusePoseInference from pose_align import PoseAlignmentInference from downloading_weights import download_models class App: def __init__(self, args): self.args = args self.pose_alignment_infer = PoseAlignmentInference( model_dir=args.model_dir, output_dir=args.output_dir ) self.musepose_infer = MusePoseInference( model_dir=args.model_dir, output_dir=args.output_dir ) if not args.disable_model_download_at_start: download_models(model_dir=args.model_dir) @staticmethod def on_step1_complete(input_img: str, input_pose_vid: str): return [gr.Image(label="Input Image", value=input_img, type="filepath", scale=5), gr.Video(label="Input Aligned Pose Video", value=input_pose_vid, scale=5)] def musepose_demo(self): with gr.Blocks() as demo: md_header = self.header() with gr.Tabs(): with gr.TabItem('1: Pose Alignment'): with gr.Row(): with gr.Column(scale=3): img_pose_input = gr.Image(label="Input Image", type="filepath", scale=5) vid_dance_input = gr.Video(label="Input Dance Video", max_length=10, scale=5) # Changed max_length to 10 with gr.Column(scale=3): vid_dance_output = gr.Video(label="Aligned Pose Output", scale=5, interactive=False) vid_dance_output_demo = gr.Video(label="Aligned Pose Output Demo", scale=5) # Rest of the column setup remains the same with gr.Column(scale=3): # column settings remain the same # button settings and event handlers remain the same with gr.TabItem('2: MusePose Inference'): with gr.Row(): with gr.Column(scale=3): img_musepose_input = gr.Image(label="Input Image", type="filepath", scale=5) vid_pose_input = gr.Video(label="Input Aligned Pose Video", max_length=10, scale=5) # Changed max_length to 10 with gr.Column(scale=3): vid_output = gr.Video(label="MusePose Output", scale=5) vid_output_demo = gr.Video(label="MusePose Output Demo", scale=5) # Rest of the settings remains the same # Event handler and button settings remain unchanged return demo @staticmethod def header(): header = gr.HTML( """

MusePose WebUI

Note: This space now allows video input up to 10 seconds because ZeroGPU limits the function runtime to 2 minutes.
If you want longer video inputs, you have to run it locally. Click the link above and follow the README to try it locally.

When you have completed the 1: Pose Alignment process, go to 2: MusePose Inference and click the "GENERATE" button.

""" ) return header def launch(self): demo = self.musepose_demo() demo.queue().launch( share=self.args.share ) if __name__ == "__main__": parser = argparse.ArgumentParser() parser.add_argument('--model_dir', type=str, default=os.path.join("pretrained_weights"), help='Pretrained models directory for MusePose') parser.add_argument('--output_dir', type=str, default=os.path.join("outputs"), help='Output directory for the result') parser.add_argument('--disable_model_download_at_start', type=bool, default=False, nargs='?', const=True, help='Disable model download at start or not') parser.add_argument('--share', type=bool, default=False, nargs='?', const=True, help='Gradio makes sharable link if it is true') args = parser.parse_args() app = App(args=args) app.launch()