fix: chunked VAE encoding for OOM prevention

src/scope/core/pipelines/wan2_1/blocks/prepare_video_latents.py

@@ -73,13 +73,32 @@ def intermediate_outputs(self) -> list[OutputParam]:
             OutputParam("generator", description="Random number generator"),
         ]
 
+    # Maximum number of frames to encode at once to avoid CUDA OOM.
+    VAE_ENCODE_CHUNK_SIZE = 8
+
     @torch.no_grad()
     def __call__(self, components, state: PipelineState) -> tuple[Any, PipelineState]:
         block_state = self.get_block_state(state)
 
-        # Encode frames to latents using VAE
-        # VAE returns [B, F, C, H, W] which is what DenoiseBlock/Generator expect
-        latents = components.vae.encode_to_latent(block_state.video)
+        # Encode frames to latents using VAE in chunks to prevent CUDA OOM.
+        # VAE expects [B, C, F, H, W] and returns [B, F, C, H, W].
+        video = block_state.video
+        # video shape: [B, C, F, H, W]
+        num_frames = video.shape[2]
+        chunk_size = self.VAE_ENCODE_CHUNK_SIZE
+
+        if num_frames <= chunk_size:
+            latents = components.vae.encode_to_latent(video)
+        else:
+            latent_chunks = []
+            for start in range(0, num_frames, chunk_size):
+                end = min(start + chunk_size, num_frames)
+                chunk = video[:, :, start:end, :, :]
+                latent_chunk = components.vae.encode_to_latent(chunk)
+                latent_chunks.append(latent_chunk)
+                torch.cuda.empty_cache()
+            # Concatenate along frames dim (dim=1 in latent space [B, F, C, H, W])
+            latents = torch.cat(latent_chunks, dim=1)
 
         # The default param for InputParam does not work right now
         # The workaround is to set the default values here