NVIDIA Simplifies Video Generation with ComfyUI for AI

At the Game Developers Conference (GDC) in San Francisco, NVIDIA unveiled updates that simplify video generation using AI for concept development and storyboarding on RTX graphics processors and the NVIDIA DGX Spark supercomputer.

Among the new features is a new App View mode in ComfyUI with a streamlined interface, lowering the entry barrier for the popular generative AI tool. RTX Video Super Resolution is also available for ComfyUI, enabling real-time video upscaling to 4K.

For FLUX.2 Klein, NVFP4 and FP8 model options are available, providing up to 2.5x performance gains and 60% less memory usage. ComfyUI now natively supports these data formats, and performance for RTX graphics processors has increased by 40% since September.

Users can quickly generate 4K video using NVIDIA RTX Video Super Resolution, available as a standalone node for video workflow creation. NVIDIA has also released a free Python package for AI developers.

This package provides programmatic access to the video upscaling technology running on the RTX graphics processors' Tensor Cores, allowing for 4K output 30 times faster than alternative local upscalers.

To get started with ComfyUI, users can download the NVFP4 and FP8 checkpoints directly from Hugging Face and load workflows through the Template Browser.

The ComfyUI updates are available today, and users can check out the latest NVIDIA Studio Sessions tutorials for more detailed information.