Dataset · TempoFunk
Video latent representations and CLIP prompt embeddings
TempoFunk/small
7.8k videos as latent arrays and CLIP prompt embeddings for video generation.
What's inside
The set includes 7,800 examples of random English-language videos already encoded into latent space. Each sample consists of latents for 120 frames of original 512x512 size in NumPy array format (shape 120x4x64x64) and text descriptions encoded via OpenAI CLIP (shape 77x768). Additionally, JSON metadata with descriptions, tags, categories, and source links are included.
What you can do with it
- Training and fine-tuning text-to-video models without VAE encoding overhead
- Training video and image classifiers on latent features
- Experiments with generative diffusion architectures
- Tasks
- Text → video, Text → image, Video classification, Image classification
- Languages
- English
- Size
- 1K–10K rows
- Downloads this month
- 22.4K
- Likes
- 13
How to load it
from datasets import load_dataset
ds = load_dataset("TempoFunk/small")