ComfyUI Tutorial Series Ep 51: Nvidia Cosmos Predict2 Image & Video Models in Action

Sobre esta lección

In episode 51 of the ComfyUI tutorial series, we take a detailed look at NVIDIA Cosmos Predict 2 models for both image and video generation. Learn how to set up the workflow in ComfyUI, download and organize the required models, and generate high-quality outputs using the 2B base version or larger GGUF variants like Q4 and Q8. This episode covers both text-to-image and image-to-image workflows, how to manage denoise settings, and how to use helper tools like resolution calculators and image scaling nodes for better results. We also test out the video generation workflow, starting from an image and creating short video clips using the "video-to-world" node. You'll see side-by-side comparisons of Cosmos 2B, 14B Q4/Q8, and popular models like Flux Dev, along with insights on which model performs better for different styles — including anime, oil painting, cartoon renders, and realism. 📌 What you’ll learn: - How to update ComfyUI for Cosmos support - Setup for text-to-image and image-to-image workflows - Using Cosmos models in GGUF format - Running the video generation workflow - Model comparison: Cosmos vs Flux for different styles 📂 Workflows mentioned in the video are available for free download on the Pixaroma Discord. Here - https://discord.gg/gggpkVgBf3 After you joined discord you can access this direct link to find all the workflows for this video https://discord.com/channels/1245221993746399232/1387063560869384222/1387063863832477766 Check Other Episodes https://www.youtube.com/playlist?list=PL-pohOSaL8P9kLZP8tQ1K1QWdZEgwiBM0 Unlock exclusive perks by joining our channel: https://www.youtube.com/channel/UCmMbwA-s3GZDKVzGZ-kPwaQ/join Cosmos Models https://huggingface.co/Comfy-Org/Cosmos_Predict2_repackaged/tree/main GGUF Version Models https://huggingface.co/calcuis/cosmos-predict2-gguf/tree/main Download oldt5_xxl_fp8_e4m3fn_scaled.safetensors from https://huggingface.co/comfyanonymous/cosmos_1.0_text_encoder_and_VAE_ComfyUI/resolve/main/text_encoders/oldt5_xxl_fp8_e4m3fn_scaled.safetensors Download wan_2.1_vae.safetensors model from https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/vae/wan_2.1_vae.safetensors Read more on the Comfy Blog https://blog.comfy.org/p/cosmos-predict2-now-supported-in Cosmos Documentation https://docs.comfy.org/tutorials/image/cosmos/cosmos-predict2-t2i #comfyui #comfyuitutorial #imagegenerator Chapters 00:00 Introduction 00:14 Cosmos 2B text2img workflow setup 04:11 Cosmos 2B img2img workflow setup 07:07 Cosmos 14B GGUF Workflow 09:16 Cosmos 2B vs 14B Q4 GGUF vs 14B Q8 GGUF 11:11 Cosmos 14B Q8 vs Flux Dev Q8 13:01 Cosmos Video model - video2world