t2v, i2v outputs black screen

[eisneim]

follow the instructions in the readme.md i can run inference without error, but got black screen video, after some inspection, the output of
images = pipeline is [ nan, nan, nan, nan...... ]

pytorch: 2.3.1+cu118
diffusers: 0.32.0
transformers: 4.47.1

WARNING:__main__:Running generation with arguments: Namespace(ckpt_path='./pretrained/ltx-video-2b-v0.9.1.safetensors', input_video_path=None, input_image_path=None, output_path=None, seed=12345, num_inference_steps=40, num_images_per_prompt=1, guidance_scale=3, stg_scale=1, stg_rescale=0.7, stg_mode='stg_a', stg_skip_layers='19', image_cond_noise_scale=0.15, height=512, width=768, num_frames=121, frame_rate=25, precision='bfloat16', decode_timestep=0.05, decode_noise_scale=0.025, prompt='a chinese woman lying on a field of lotus, she is blinking her eyes, light wind blowing, no camera movement', negative_prompt='worst quality, inconsistent motion, blurry, jittery, distorted', offload_to_cpu=False)
WARNING:__main__:Padded dimensions: 512x768x121
Downloading shards: 100%|██████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 16194.22it/s]
Loading checkpoint shards: 100%|██████████████████████████████████████████████████████| 2/2 [00:00<00:00,  9.37it/s]
  0%|                                                                                        | 0/40 [00:00<?, ?it/s]/home/eisneim/.conda/envs/_learn/lib/python3.10/site-packages/torch/functional.py:512: UserWarning: torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument. (Triggered internally at ../aten/src/ATen/native/TensorShape.cpp:3587.)
  return _VF.meshgrid(tensors, **kwargs)  # type: ignore[attr-defined]
100%|███████████████████████████████████████████████████████████████████████████████| 40/40 [00:33<00:00,  1.19it/s]
/home/eisneim/www/ml/video_gen/LTX-Video/ltx_video/pipelines/pipeline_ltx_video.py:1172: FutureWarning: Accessing config attribute `in_channels` directly via 'Transformer3DModel' object attribute is deprecated. Please access 'in_channels' over 'Transformer3DModel's config object instead, e.g. 'unet.config.in_channels'.
  out_channels=self.transformer.in_channels
/home/eisneim/www/ml/video_gen/LTX-Video/inference.py:429: RuntimeWarning: invalid value encountered in cast
  video_np = (video_np * 255).astype(np.uint8)
WARNING:__main__:Output saved to outputs/2024-12-25

strangely, on another ubuntu 22.04 with pytorch 2.5 it can successfully generate videos with same code and prompt

[andreclaudino]

I am getting black screen too. When using comfyUI it works, but when using inference script it returns a black screen.