README.md

metadata

license: apache-2.0
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
dataset_info:
  features:
    - name: contentUrl
      dtype: string
    - name: name
      dtype: string
    - name: dynamic_source_category
      dtype: string
    - name: video_tensor
      sequence:
        sequence:
          sequence:
            sequence: float32
    - name: shape
      sequence: int64
    - name: hyv_latents
      sequence:
        sequence:
          sequence:
            sequence: float32
    - name: hyv_latent_shape
      sequence: int64
    - name: text_encoded
      sequence:
        sequence:
          sequence: float32
  splits:
    - name: train
      num_bytes: 795314833
      num_examples: 64
  download_size: 196482084
  dataset_size: 795314833
task_categories:
  - text-to-video
language:
  - en
tags:
  - video
  - preprocessed
size_categories:
  - 1K<n<10K

tiny_webvid_latents

This is a tiny processed split of Doubiiu/webvid10m_motion.

• resized to 224x224
• Trimmed to 5s and 4fps
• Encoded to video latents with hunyuanvideo-community/HunyuanVideo's video/3D VAE
• text/caption encoded with google-t5/t5-base