[Misc] Fix memory leakage in VideoGenerator (#1132)

README.md

@@ -93,7 +93,6 @@ def main():
     # Generate the video
     video = generator.generate_video(
         prompt,
-        return_frames=True,  # Also return frames from this call (defaults to False)
         output_path="my_videos/",  # Controls where videos are saved
         save_video=True
     )

docs/getting_started/quick_start.md

@@ -41,7 +41,6 @@ def main():
     # Generate the video
     video = generator.generate_video(
         prompt,
-        return_frames=True,  # Also return frames from this call (defaults to False)
         output_path="my_videos/",  # Controls where videos are saved
         save_video=True
     )

docs/inference/configuration.md

@@ -61,12 +61,11 @@ def main():
         prompt, 
         sampling_param=sampling_param, 
         output_path="my_videos/",  # Controls where videos are saved
-        return_frames=True,  # Also return frames from this call (defaults to False)
         save_video=True
     )
 
-    # If return_frames=True, video contains the generated frames as a NumPy array
-    print(f"Generated {len(video)} frames")
+    # If return_frames=True, frames are available in video["frames"]
+    print(f"Generated {len(video['frames'])} frames")
 
 if __name__ == '__main__':
     main()
@@ -76,6 +75,8 @@ if __name__ == '__main__':
 
 The CLI supports `--config` with JSON or YAML. Command-line arguments override
 config file values.
+By default, `fastvideo generate` uses `return_frames=false` unless you set
+`--return-frames` (or `return_frames: true` in config).
 
 ```bash
 fastvideo generate --config config.yaml

docs/inference/inference_quick_start.md

@@ -44,7 +44,6 @@ def main():
     # Generate the video
     video = generator.generate_video(
         prompt,
-        return_frames=True,  # Also return frames from this call (defaults to False)
         output_path="my_videos/",  # Controls where videos are saved
         save_video=True
     )

examples/inference/gradio/serving/ray_serve_backend.py

@@ -168,7 +168,7 @@ def prepare_sampling_params(video_request: VideoGenerationRequest, default_param
     params.height = video_request.height
     params.width = video_request.width
     params.save_video = False
-    params.return_frames = False
+    params.return_frames = True
     
     return params
 
@@ -229,7 +229,7 @@ class BaseModelDeployment:
             sampling_param=params,
             image_path=image_path,
             save_video=False,
-            return_frames=False,
+            return_frames=True,
         )
         inference_time = time.time() - inference_start_time
 

fastvideo/configs/sample/base.py

@@ -75,7 +75,7 @@ class SamplingParam:
 
     # Misc
     save_video: bool = True
-    return_frames: bool = False
+    return_frames: bool = True
     return_trajectory_latents: bool = False  # returns all latents for each timestep
     return_trajectory_decoded: bool = False  # returns decoded latents for each timestep
 
@@ -218,7 +218,7 @@ class SamplingParam:
         parser.add_argument(
             "--return-frames",
             action="store_true",
-            default=SamplingParam.return_frames,
+            default=False,
             help="Whether to return the raw frames",
         )
         parser.add_argument(

fastvideo/entrypoints/cli/generate.py

@@ -80,6 +80,7 @@ class GenerateSubcommand(CLISubcommand):
             k: v
             for k, v in merged_args.items() if k in self.generation_arg_names
         }
+        generation_args.setdefault("return_frames", False)
 
         model_path = init_args.pop('model_path')
         prompt = generation_args.pop('prompt', None)

fastvideo/entrypoints/video_generator.py

@@ -125,7 +125,7 @@ class VideoGenerator:
         grid_sizes: tuple[int, int, int] | list[int] | torch.Tensor
         | None = None,
         **kwargs,
-    ) -> dict[str, Any] | list[np.ndarray] | list[dict[str, Any]]:
+    ) -> dict[str, Any] | list[dict[str, Any]]:
         """
         Generate a video based on the given prompt.
         
@@ -135,7 +135,7 @@ class VideoGenerator:
             output_path: Path to save the video (overrides the one in fastvideo_args)
             prompt_path: Path to prompt file
             save_video: Whether to save the video to disk
-            return_frames: Whether to return the raw frames
+            return_frames: Whether to include raw frames in the result dict
             num_inference_steps: Number of denoising steps (overrides fastvideo_args)
             guidance_scale: Classifier-free guidance scale (overrides fastvideo_args)
             num_frames: Number of frames to generate (overrides fastvideo_args)
@@ -147,7 +147,8 @@ class VideoGenerator:
             callback_steps: Number of steps between each callback
             
         Returns:
-            Either the output dictionary, list of frames, or list of results for batch processing
+            A metadata dictionary for single-prompt generation, or a list of
+            metadata dictionaries for prompt-file batch generation.
         """
         # Handle batch processing from text file
         if sampling_param is None:
@@ -194,9 +195,8 @@ class VideoGenerator:
                         **kwargs)
 
                     # Add prompt info to result
-                    if isinstance(result, dict):
-                        result["prompt_index"] = i
-                        result["prompt"] = batch_prompt
+                    result["prompt_index"] = i
+                    result["prompt"] = batch_prompt
 
                     results.append(result)
                     logger.info("Successfully generated video for prompt %d",
@@ -306,7 +306,7 @@ class VideoGenerator:
         prompt: str,
         sampling_param: SamplingParam | None = None,
         **kwargs,
-    ) -> dict[str, Any] | list[np.ndarray]:
+    ) -> dict[str, Any]:
         """Internal method for single video generation"""
         # Create a copy of inference args to avoid modifying the original
         fastvideo_args = self.fastvideo_args
@@ -475,21 +475,22 @@ class VideoGenerator:
                     logger.warning(
                         "Audio mux failed; saved video without audio.")
 
-        if batch.return_frames:
-            return frames
-        else:
-            return {
-                "samples": samples,
-                "frames": frames,
-                "audio": output_batch.extra.get("audio"),
-                "prompts": prompt,
-                "size": (target_height, target_width, batch.num_frames),
-                "generation_time": gen_time,
-                "logging_info": logging_info,
-                "trajectory": output_batch.trajectory_latents,
-                "trajectory_timesteps": output_batch.trajectory_timesteps,
-                "trajectory_decoded": output_batch.trajectory_decoded,
-            }
+        result: dict[str, Any] = {
+            "prompts": prompt,
+            "samples": samples if batch.return_frames else None,
+            "frames": frames if batch.return_frames else None,
+            "audio":
+            output_batch.extra.get("audio") if batch.return_frames else None,
+            "size": (target_height, target_width, batch.num_frames),
+            "generation_time": gen_time,
+            "logging_info": logging_info,
+            "trajectory": output_batch.trajectory_latents,
+            "trajectory_timesteps": output_batch.trajectory_timesteps,
+            "trajectory_decoded": output_batch.trajectory_decoded,
+            "video_path": output_path if batch.save_video else None,
+        }
+
+        return result
 
     @staticmethod
     def _mux_audio(

tests/local_tests/pipelines/test_gamecraft_pipeline_parity.py

@@ -340,9 +340,8 @@ def test_gamecraft_pipeline_latent_parity():
         camera_states=plucker_embedding,
     )
     
-    fastvideo_latents = result.get("latents")
-    if fastvideo_latents is None:
-        fastvideo_latents = result.get("samples")
+
+    fastvideo_latents = result.get("samples")
     
     _log_tensor_stats("FastVideo latents", fastvideo_latents)