Show-1https://showlab.github.io/Show-1/

Show-1

šŸš€ Introducing Show-1: the cutting-edge AI tool that combines the best of both worlds - pixel and latent diffusion models for remarkable text-to-video generation! šŸŽ„šŸ¤–āœØ āœØ Achieve precise text-video alignment with reduced computational costs. šŸŒŸ Available for public use! #AI #VideoGeneration #Show1

  • Show-1 is a hybrid model that combines pixel-based and latent-based VDMs for text-to-video generation.
  • Pixel-based VDMs offer accurate motion aligned with text but require high computational costs.
  • Latent-based VDMs are more resource-efficient but struggle with precise text-video alignment due to small latent space.
  • Show-1 first uses pixel-based VDMs for low-resolution video with strong text-video correlation.
  • It then employs a novel expert translation method using latent-based VDMs to upscale to high resolution.
  • Show-1 balances quality and efficiency: precise alignment like latent VDMs and reduced GPU memory usage like pixel VDMs.
  • The model is validated on standard video generation benchmarks and is publicly available.