CogVideoX is a family of AI models that generate coherent, high-quality videos from text prompts by employing a 3D Variational Autoencoder for spatial and temporal compression, an expert transformer for enhanced text-video alignment, and progressive training techniques to produce longer, dynamic videos with significant motion.

AuraFlow is an open-source family of text-to-image models developed by fal.ai that employs flow-based generative architecture rather than diffusion methods. The models feature modified MMDiT architecture with enhanced computational efficiency and utilize curated training datasets with custom caption systems to improve prompt-image alignment. The family includes iterative releases from v0.1 to v0.3, with each version incorporating architectural improvements and better scaling methodologies for research and creative applications.