Upgraded gradio to 4.43, hoping their connectivity issue is fixed

Added new action when no face detected -> use last swapped
Specified image format for image controls - opening new tabs on preview images possible again!
Hardcoded image output format for livecam to jpeg - might be faster than previous webp
Chain events to be only executed if previous was a success

requirements.txt

@@ -1,7 +1,7 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
 
 numpy==1.26.4
-gradio==4.42.0
+gradio==4.43.0
 fastapi<0.113.0
 opencv-python==4.9.0.80
 onnx==1.16.0

roop/ProcessMgr.py

@@ -26,7 +26,8 @@ class eNoFaceAction():
     USE_ORIGINAL_FRAME = 0
     RETRY_ROTATED = 1
     SKIP_FRAME = 2
-    SKIP_FRAME_IF_DISSIMILAR = 3
+    SKIP_FRAME_IF_DISSIMILAR = 3,
+    USE_LAST_SWAPPED = 4
 
 
 
@@ -70,7 +71,8 @@ class ProcessMgr():
     progress_gradio = None
     total_frames = 0
 
-
+    num_frames_no_face = 0
+    last_swapped_frame = None
 
 
     plugins =  { 
@@ -103,6 +105,8 @@ def reuseOldProcessor(self, name:str):
     def initialize(self, input_faces, target_faces, options):
         self.input_face_datas = input_faces
         self.target_face_datas = target_faces
+        self.num_frames_no_face = 0
+        self.last_swapped_frame = None
         self.options = options
         devicename = get_device()
 
@@ -329,8 +333,16 @@ def process_frame(self, frame:Frame):
             if roop.globals.no_face_action == eNoFaceAction.SKIP_FRAME_IF_DISSIMILAR:
                 if len(self.input_face_datas) > num_swapped:
                     return None
+            self.num_frames_no_face = 0
+            self.last_swapped_frame = temp_frame.copy()
             return temp_frame
-        if roop.globals.no_face_action == eNoFaceAction.USE_ORIGINAL_FRAME:
+        if roop.globals.no_face_action == eNoFaceAction.USE_LAST_SWAPPED:
+            if self.last_swapped_frame is not None and self.num_frames_no_face < self.options.max_num_reuse_frame:
+                self.num_frames_no_face += 1
+                return self.last_swapped_frame.copy()
+            return frame
+
+        elif roop.globals.no_face_action == eNoFaceAction.USE_ORIGINAL_FRAME:
             return frame
         if roop.globals.no_face_action == eNoFaceAction.SKIP_FRAME:
             #This only works with in-mem processing, as it simply skips the frame.

roop/ProcessOptions.py

@@ -12,4 +12,5 @@ def __init__(self, processordefines:dict, face_distance,  blend_ratio, swap_mode
         self.show_face_area_overlay = show_face_area
         self.show_face_masking = show_mask
         self.subsample_size = subsample_size
-        self.restore_original_mouth = restore_original_mouth
+        self.restore_original_mouth = restore_original_mouth
+        self.max_num_reuse_frame = 15

roop/metadata.py

@@ -1,2 +1,2 @@
 name = 'roop unleashed'
-version = '4.2.1'
+version = '4.2.3'

ui/tabs/facemgr_tab.py

@@ -22,7 +22,7 @@ def facemgr_tab() -> None:
                         Add multiple reference images into a faceset file.
                         """)
         with gr.Row():
-            videoimagefst = gr.Image(label="Cut face from video frame", height=576, interactive=False, visible=True)
+            videoimagefst = gr.Image(label="Cut face from video frame", height=576, interactive=False, visible=True, format="jpeg")
         with gr.Row():
             frame_num_fst = gr.Slider(1, 1, value=1, label="Frame Number", info='0:00:00', step=1.0, interactive=False)
             fb_cutfromframe = gr.Button("Use faces from this frame", variant='secondary', interactive=False)

ui/tabs/faceswap_tab.py

@@ -30,7 +30,7 @@
 is_processing = False            
 
 list_files_process : list[ProcessEntry] = []
-no_face_choices = ["Use untouched original frame","Retry rotated", "Skip Frame", "Skip Frame if no similar face"]
+no_face_choices = ["Use untouched original frame","Retry rotated", "Skip Frame", "Skip Frame if no similar face", "Use last swapped"]
 
 current_video_fps = 50
 
@@ -144,7 +144,7 @@ def faceswap_tab():
                     forced_fps = gr.Slider(minimum=0, maximum=120, value=0, label="Video FPS", info='Overrides detected fps if not 0', step=1.0, interactive=True, container=True)
 
             with gr.Column(scale=2):
-                previewimage = gr.Image(label="Preview Image", height=576, interactive=False, visible=True)
+                previewimage = gr.Image(label="Preview Image", height=576, interactive=False, visible=True, format=get_gradio_output_format())
                 maskimage = gr.ImageEditor(label="Manual mask Image", sources=["clipboard"], transforms="", type="numpy",
                                              brush=gr.Brush(color_mode="fixed", colors=["rgba(255, 255, 255, 1"]), interactive=True, visible=False)
                 with gr.Row(variant='panel'):
@@ -212,7 +212,7 @@ def faceswap_tab():
     previewinputs = [preview_frame_num, bt_destfiles, fake_preview, ui.globals.ui_selected_enhancer, selected_face_detection,
                         max_face_distance, ui.globals.ui_blend_ratio, selected_mask_engine, clip_text, no_face_action, vr_mode, autorotate, maskimage, chk_showmaskoffsets, chk_restoreoriginalmouth, num_swap_steps, ui.globals.ui_upscale]
     previewoutputs = [previewimage, maskimage, preview_frame_num] 
-    input_faces.select(on_select_input_face, None, None).then(fn=on_preview_frame_changed, inputs=previewinputs, outputs=previewoutputs)
+    input_faces.select(on_select_input_face, None, None).success(fn=on_preview_frame_changed, inputs=previewinputs, outputs=previewoutputs)
 
     bt_move_left_input.click(fn=move_selected_input, inputs=[bt_move_left_input], outputs=[input_faces])
     bt_move_right_input.click(fn=move_selected_input, inputs=[bt_move_right_input], outputs=[input_faces])
@@ -234,8 +234,8 @@ def faceswap_tab():
     bt_remove_selected_target_face.click(fn=remove_selected_target_face, outputs=[target_faces])
 
     forced_fps.change(fn=on_fps_changed, inputs=[forced_fps], show_progress='hidden')
-    bt_destfiles.change(fn=on_destfiles_changed, inputs=[bt_destfiles], outputs=[preview_frame_num, text_frame_clip], show_progress='hidden').then(fn=on_preview_frame_changed, inputs=previewinputs, outputs=previewoutputs, show_progress='hidden')
-    bt_destfiles.select(fn=on_destfiles_selected, outputs=[preview_frame_num, text_frame_clip, forced_fps], show_progress='hidden').then(fn=on_preview_frame_changed, inputs=previewinputs, outputs=previewoutputs, show_progress='hidden')
+    bt_destfiles.change(fn=on_destfiles_changed, inputs=[bt_destfiles], outputs=[preview_frame_num, text_frame_clip], show_progress='hidden').success(fn=on_preview_frame_changed, inputs=previewinputs, outputs=previewoutputs, show_progress='hidden')
+    bt_destfiles.select(fn=on_destfiles_selected, outputs=[preview_frame_num, text_frame_clip, forced_fps], show_progress='hidden').success(fn=on_preview_frame_changed, inputs=previewinputs, outputs=previewoutputs, show_progress='hidden')
     bt_destfiles.clear(fn=on_clear_destfiles, outputs=[target_faces, selected_face_detection])
     resultfiles.select(fn=on_resultfiles_selected, inputs=[resultfiles], outputs=[resultimage, resultvideo])
 
@@ -252,7 +252,7 @@ def faceswap_tab():
         inputs=[ui.globals.ui_selected_enhancer, selected_face_detection, roop.globals.keep_frames, roop.globals.wait_after_extraction,
                     roop.globals.skip_audio, max_face_distance, ui.globals.ui_blend_ratio, selected_mask_engine, clip_text,video_swapping_method, no_face_action, vr_mode, autorotate, chk_restoreoriginalmouth, num_swap_steps, ui.globals.ui_upscale, maskimage],
         outputs=[bt_start, bt_stop, resultfiles], show_progress='full')
-    after_swap_event = start_event.then(fn=on_resultfiles_finished, inputs=[resultfiles], outputs=[resultimage, resultvideo])
+    after_swap_event = start_event.success(fn=on_resultfiles_finished, inputs=[resultfiles], outputs=[resultimage, resultvideo])
 
     bt_stop.click(fn=stop_swap, cancels=[start_event, after_swap_event], outputs=[bt_start, bt_stop], queue=False)
 
@@ -813,6 +813,12 @@ def on_resultfiles_finished(files):
     return display_output(filename)
 
 
+def get_gradio_output_format():
+    if roop.globals.CFG.output_image_format == "jpg":
+        return "jpeg"
+    return roop.globals.CFG.output_image_format
+
+
 def display_output(filename):
     if util.is_video(filename) and roop.globals.CFG.output_show_video:
         return gr.Image(visible=False), gr.Video(visible=True, value=filename)

ui/tabs/livecam_tab.py

@@ -27,7 +27,7 @@ def livecam_tab():
                 dd_reso = gr.Dropdown(choices=["640x480","1280x720", "1920x1080"], value="1280x720", label="Fake Camera Resolution", interactive=True)
 
         with gr.Row():
-            fake_cam_image = gr.Image(label='Fake Camera Output', interactive=False)
+            fake_cam_image = gr.Image(label='Fake Camera Output', interactive=False, format="jpeg")
 
     start_event = bt_start.click(fn=start_cam,  inputs=[cb_obs, camera_num, dd_reso, ui.globals.ui_selected_enhancer, ui.globals.ui_blend_ratio, ui.globals.ui_upscale],outputs=[bt_start, bt_stop,fake_cam_image])
     bt_stop.click(fn=stop_swap, cancels=[start_event], outputs=[bt_start, bt_stop], queue=False)