update ui and bug fix

app.py

@@ -40,7 +40,7 @@ def videocrafter_demo(result_dir='./tmp/'):
                             lora_scale = gr.Slider(minimum=0.0, maximum=2.0, step=0.1, label='Lora Scale', value=1.0, elem_id="lora_scale")
                             cfg_scale = gr.Slider(minimum=1.0, maximum=30.0, step=0.5, label='CFG Scale', value=15.0, elem_id="cfg_scale")
                         send_btn = gr.Button("Send")
-                    with gr.Tab(label='show'):
+                    with gr.Tab(label='result'):
                         output_video_1 =  gr.Video().style(width=384)
                 gr.Examples(examples=t2v_examples,
                             inputs=[input_text,steps,model_index,eta,cfg_scale,lora_scale],
@@ -59,7 +59,9 @@ def videocrafter_demo(result_dir='./tmp/'):
                 with gr.Row():
                     # with gr.Tab(label='input'):
                     with gr.Column():
-                        vc_input_video = gr.Video().style(width=256)
+                        with gr.Row():
+                            vc_input_video = gr.Video(label="Input Video").style(width=256)
+                            vc_origin_video = gr.Video(label='Center-cropped Video').style(width=256)
                         with gr.Row():
                             vc_input_text = gr.Text(label='Prompts')
                         with gr.Row():
@@ -72,16 +74,18 @@ def videocrafter_demo(result_dir='./tmp/'):
                         vc_end_btn = gr.Button("Send")
                     with gr.Tab(label='Result'):
                         vc_output_info = gr.Text(label='Info')
-                        vc_output_video = gr.Video().style(width=384)
+                        with gr.Row():
+                            vc_depth_video = gr.Video(label="Depth Video").style(width=256)
+                            vc_output_video = gr.Video(label="Generated Video").style(width=256)
 
                 gr.Examples(examples=control_examples,
                             inputs=[vc_input_video, vc_input_text, frame_stride, vc_steps, vc_cfg_scale, vc_eta],
-                            outputs=[vc_output_info, vc_output_video],
+                            outputs=[vc_output_info, vc_origin_video, vc_depth_video, vc_output_video],
                             fn = videocontrol.get_video,
-                            cache_examples=False
+                            cache_examples=os.getenv('SYSTEM') == 'spaces',
                 )
             vc_end_btn.click(inputs=[vc_input_video, vc_input_text, frame_stride, vc_steps, vc_cfg_scale, vc_eta],
-                            outputs=[vc_output_info, vc_output_video],
+                            outputs=[vc_output_info, vc_origin_video, vc_depth_video, vc_output_video],
                             fn = videocontrol.get_video
             )
 

videocontrol_test.py

@@ -70,7 +70,7 @@ class VideoControl:
             h, w, c = VideoReader(input_video, ctx=cpu(0))[0].shape
         except:
             os.remove(input_video)
-            return 'please input video', None
+            return 'please input video', None, None, None
 
         if h < w:
             scale = h / self.resolution
@@ -82,7 +82,7 @@ class VideoControl:
             video, info_str = load_video(input_video, frame_stride, video_size=(h, w), video_frames=16)
         except:
             os.remove(input_video)
-            return 'load video error', None 
+            return 'load video error', None, None, None
         video = self.spatial_transform(video)
         print('video shape', video.shape)
 
@@ -103,14 +103,23 @@ class VideoControl:
         filename = prompt
         filename = filename.replace("/", "_slash_") if "/" in filename else filename
         filename = filename.replace(" ", "_") if " " in filename else filename
+        if len(filename) > 200:
+            filename = filename[:200]
         video_path = os.path.join(self.savedir, f'{filename}_sample.mp4')
-        # tensor_to_mp4(video=batch_conds.detach().cpu(), savepath=os.path.join(self.savedir, f'{filename}_depth.mp4'), fps=10)
-        tensor_to_mp4(video=batch_samples.detach().cpu(), savepath=os.path.join(self.savedir, f'{filename}_sample.mp4'), fps=8)
+        depth_path = os.path.join(self.savedir, f'{filename}_depth.mp4')
+        origin_path = os.path.join(self.savedir, f'{filename}.mp4')
+        tensor_to_mp4(video=video.detach().cpu(), savepath=origin_path, fps=8)
+        tensor_to_mp4(video=batch_conds.detach().cpu(), savepath=depth_path, fps=8)
+        tensor_to_mp4(video=batch_samples.detach().cpu(), savepath=video_path, fps=8)
 
         print(f"Saved in {video_path}. Time used: {(time.time() - start):.2f} seconds")
         # delete video
-        os.remove(input_video)
-        return info_str, video_path
+        (path, input_filename) = os.path.split(input_video)
+        if input_filename != 'flamingo.mp4':
+            os.remove(input_video)
+            print('delete input video')
+        # print(input_video)
+        return info_str, origin_path, depth_path, video_path
     def download_model(self):
         REPO_ID = 'VideoCrafter/t2v-version-1-1'
         filename_list = ['models/base_t2v/model.ckpt',