Overriding the new default of 'sdpa'

.gitignore

@@ -0,0 +1,7 @@
+__pycache__/
+weights/
+local_data/
+save_audio/
+.vscode/
+*.mp4
+.venv/

app.py

@@ -21,7 +21,7 @@
 import wan
 from wan.configs import SIZE_CONFIGS, SUPPORTED_SIZES, WAN_CONFIGS
 from wan.utils.utils import cache_image, cache_video, str2bool
-from wan.utils.multitalk_utils import save_video_ffmpeg
+from wan.utils.multitalk_utils import ChunkedVideoWriter, mux_video_with_audio
 from kokoro import KPipeline
 from transformers import Wav2Vec2FeatureExtractor
 from src.audio_analysis.wav2vec2 import Wav2Vec2Model
@@ -602,33 +602,47 @@ def generate_video(img2vid_image, img2vid_prompt, n_prompt, img2vid_audio_1, img
         #     input_data['cond_audio']['person1'] = emb_path
         #     input_data['video_audio'] = sum_audio
 
-        logging.info("Generating video ...")
-        video = wan_i2v.generate(
-            input_data,
-            size_buckget=resolution_select,
-            motion_frame=args.motion_frame,
-            frame_num=args.frame_num,
-            shift=args.sample_shift,
-            sampling_steps=sd_steps,
-            text_guide_scale=text_guide_scale,
-            audio_guide_scale=audio_guide_scale,
-            seed=seed,
-            n_prompt=n_prompt,
-            offload_model=args.offload_model,
-            max_frames_num=args.frame_num if args.mode == 'clip' else 1000,
-            color_correction_strength = args.color_correction_strength,
-            extra_args=args,
-            )
-
-
         if args.save_file is None:
             formatted_time = datetime.now().strftime("%Y%m%d_%H%M%S")
             formatted_prompt = input_data['prompt'].replace(" ", "_").replace("/",
                                                                         "_")[:50]
             args.save_file = f"{args.task}_{args.size.replace('*','x') if sys.platform=='win32' else args.size}_{args.ulysses_size}_{args.ring_size}_{formatted_prompt}_{formatted_time}"
-
-        logging.info(f"Saving generated video to {args.save_file}.mp4")
-        save_video_ffmpeg(video, args.save_file, [input_data['video_audio']], high_quality_save=False)
+
+        save_path_tmp = f"{args.save_file}-temp.mp4"
+        video_writer = ChunkedVideoWriter(save_path_tmp)
+
+        logging.info("Generating video ...")
+        try:
+            frames_written = wan_i2v.generate(
+                input_data,
+                size_buckget=resolution_select,
+                motion_frame=args.motion_frame,
+                frame_num=args.frame_num,
+                shift=args.sample_shift,
+                sampling_steps=sd_steps,
+                text_guide_scale=text_guide_scale,
+                audio_guide_scale=audio_guide_scale,
+                seed=seed,
+                n_prompt=n_prompt,
+                offload_model=args.offload_model,
+                max_frames_num=args.frame_num if args.mode == 'clip' else 1000,
+                color_correction_strength = args.color_correction_strength,
+                extra_args=args,
+                video_writer=video_writer,
+                )
+        finally:
+            video_writer.close()
+
+        frame_count = frames_written if frames_written is not None else video_writer.frames_written
+        logging.info(f"Saving generated video to {args.save_file}.mp4 ({frame_count} frames)")
+        mux_video_with_audio(
+            save_path_tmp=save_path_tmp,
+            save_path=args.save_file,
+            vocal_audio_list=[input_data['video_audio']],
+            frame_count=frame_count,
+            fps=25,
+            high_quality_save=False,
+        )
         logging.info("Finished.")
 
         return args.save_file + '.mp4'

backend_runs/69a0754e42141599e5301cff/f799126dcf204820adca9bfdf0a35b79.json

@@ -0,0 +1 @@
+{"prompt": "A professional speaks confidently directly to the camera.", "cond_image": "/mnt/c/Users/anwan/OneDrive/Khan/maity/vidLink/output/videos/69a0754e42141599e5301cff/avatar.png", "tts_audio": {"text": "We are ready for liftoff... finally!", "human1_voice": "/mnt/c/Users/anwan/OneDrive/Khan/maity/vidLink/video_generators/multitalk/weights/Kokoro-82M/voices/af_heart.pt"}, "cond_audio": {}}

base_tts_template.json

@@ -0,0 +1,10 @@
+{
+    "prompt": "A person speaks to the camera.",
+    "cond_image": "Input_outputs/input_files/sales_executive/executive.png",
+    "tts_audio": {
+        "text": "example",
+        "human1_voice": "weights/Kokoro-82M/voices/af_heart.pt"
+    },
+    "cond_audio": {}
+}
+