ComfyUI now supports Nvidia Cosmos!
Blog post from Comfy
ComfyUI has announced support for Nvidia's latest Cosmos models, specifically the 7B and 14B text-to-video and image-to-video diffusion models, which are touted as state-of-the-art for video generation. The 7B model is recommended for most users due to its capability to run on a 24GB GPU at full precision and can also function on a 12GB GPU with automatic weight offloading. A notable feature of these models is their highly efficient video VAE, which allows encoding and decoding of high-quality 121-frame videos at 1280x704 resolution on a 12GB VRAM GPU without tiling, making them significantly more memory-efficient than previous models. However, these models require long prompts, handle resolutions no lower than 704x704, and have a preference for generating videos with exactly 121 frames, taking over 10 minutes to render on a high-end GPU like the 4090. Despite these constraints, the models are praised for their ability to create dynamic videos with movement from image inputs and are integrated with a new sampler, res_multistep, enhancing their performance.
No tracked trend matches for this post yet.
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.