Unload models

app_start_end.py

@@ -416,6 +416,12 @@ def worker(input_image, end_image, prompts, n_prompt, seed, total_second_length,
             # Combine both image embeddings or use a weighted approach
             image_encoder_last_hidden_state = (image_encoder_last_hidden_state + end_image_encoder_last_hidden_state) / 2
 
+        # Clean GPU
+        if not high_vram:
+            unload_complete_models(
+                image_encoder
+            )
+
         # Dtype
         image_encoder_last_hidden_state = image_encoder_last_hidden_state.to(transformer.dtype)
 
@@ -425,7 +431,7 @@ def worker(input_image, end_image, prompts, n_prompt, seed, total_second_length,
         rnd = torch.Generator("cpu").manual_seed(seed)
         num_frames = latent_window_size * 4 - 3
 
-        history_latents = torch.zeros(size=(1, 16, 1 + 2 + 16, height // 8, width // 8), dtype=torch.float32).cpu()
+        history_latents = torch.zeros(size=(1, 16, 1 + 2 + 16, height // 8, width // 8), dtype=torch.float32, device=cpu)
         history_pixels = None
         total_generated_latent_frames = 0
 
@@ -546,7 +552,7 @@ def worker(input_image, end_image, prompts, n_prompt, seed, total_second_length,
                 history_pixels = soft_append_bcthw(current_pixels, history_pixels, overlapped_frames)
 
             if not high_vram:
-                unload_complete_models()
+                unload_complete_models(vae)
 
             output_filename = os.path.join(outputs_folder, f'{job_id}_{total_generated_latent_frames}.mp4')
 
@@ -902,12 +908,12 @@ def process(input_image,
             resolution=640,
             total_second_length=5,
             latent_window_size=9,
-            steps=25,
+            steps=30,
             cfg=1.0,
             gs=10.0,
             rs=0.0,
             gpu_memory_preservation=6,
-            enable_preview=True,
+            enable_preview=False,
             use_teacache=False,
             mp4_crf=16,
             fps_number=30