Other optimizations

app.py

@@ -795,8 +795,9 @@ def worker_video(input_video, prompts, n_prompt, seed, batch, resolution, total_
         stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'Video processing ...'))))
 
         # 20250506 pftq: Encode video
-        start_latent, input_image_np, video_latents, fps, height, width, input_video_pixels  = video_encode(input_video, resolution, no_resize, vae, vae_batch_size=vae_batch, device=gpu)
+        start_latent, input_image_np, video_latents, fps, height, width = video_encode(input_video, resolution, no_resize, vae, vae_batch_size=vae_batch, device=gpu)[:5]
         start_latent = start_latent.to(dtype=torch.float32).cpu()
+        video_latents = video_latents.cpu()
 
         # CLIP Vision
         stream.output_queue.push(('progress', (None, '', make_progress_bar_html(0, 'CLIP Vision encoding ...'))))
@@ -905,7 +906,7 @@ def worker_video(input_video, prompts, n_prompt, seed, batch, resolution, total_
             rnd = torch.Generator("cpu").manual_seed(seed)
 
             # 20250506 pftq: Initialize history_latents with video latents
-            history_latents = video_latents.cpu()
+            history_latents = video_latents
             total_generated_latent_frames = history_latents.shape[2]
             # 20250506 pftq: Initialize history_pixels to fix UnboundLocalError
             history_pixels = None