ComfyUI Rolls Out Major Updates With NVIDIA RTX Acceleration and New AI Model Support
The popular open-source, node-based graphical interface for generative AI workflows — ComfyUI — has received a significant performance upgrade this month, bringing powerful new capabilities for content creators and developers alike.
Up to 40% Faster Performance With NVIDIA RTX
In collaboration with NVIDIA, ComfyUI now delivers up to 40% faster performance on NVIDIA RTX GPUs, a leap that surpasses the typical 20–30% gains seen with standard GPU upgrades. This improvement dramatically speeds up generative AI tasks, making real-time content creation more accessible and efficient.
Developers looking to optimize diffusion model performance in their apps can explore detailed guidance in the NVIDIA Developer Forum.
New Model Integrations in ComfyUI
Several advanced AI models are now supported in ComfyUI, expanding the creative possibilities across image, video, and 3D content generation:
- Wan 2.2 – A cutting-edge video generation model from Wan AI that offers high-quality, controllable output for text-to-video and speech-to-video tasks. GeForce RTX and NVIDIA RTX PRO GPUs are the only GPUs capable of running the large-scale Wan 2.2 (14B) models in ComfyUI with smooth performance.
- Qwen-Image – A new image generation foundation model from Alibaba, optimized for complex text rendering and precise visual editing. On the GeForce RTX 5090, it runs 7x faster than on Apple’s M3 Ultra, offering unmatched speed and accuracy for detailed content creation.
- FLUX.1 Krea [dev] – Developed by Black Forest Labs, this open-weight model is a development version of Krea 1, trained to produce realistic and diverse images with a distinct visual style. It eliminates oversaturated textures and is designed to produce “opinionated,” creative outputs. It also runs 8x faster on the RTX 5090 compared to the M3 Ultra.
- Hunyuan3D 2.1 – Another high-performance model, focusing on generating advanced 3D content for creative and industrial use cases, now available in the ComfyUI ecosystem.
NVIDIA Accelerates Diffusion Models With NIM Microservices
NVIDIA has also introduced TensorRT-optimized versions of leading diffusion models — including Stable Diffusion 3.5 and FLUX.1 Kontext — packaged as NVIDIA NIM microservices. These optimized models can be run in ComfyUI with:
- Up to 3x faster inference speeds
- 50% reduction in VRAM usage
This allows creators to work more efficiently with powerful models, even on hardware with limited memory.
NVIDIA RTX Remix Gets a Visual Upgrade
In a parallel update, NVIDIA also released a major upgrade to RTX Remix, the modding platform for remastering classic games. The latest version introduces a path-traced particle system, enabling dramatic visual enhancements — breathing new life into old titles with modern lighting, effects, and detail.
Whether you’re building immersive games, stunning visuals, or AI-driven video content, the combined updates to ComfyUI and NVIDIA RTX platforms offer a powerful foundation for faster, smarter, and more creative workflows.
