COMFYUI – Open source, node-based graphical interface for running and building generation AI workflows for content creation – published major updates over the past month, including performance improvements of up to 40% for NVIDIA RTX GPUs, and support for new AI models such as WAN 2.2, Qwen-Image, Flux.1 KREA and more. [dev] and Hunyuan3d 2.1.
Nvidia has also released the Nvidia Tensortort-Optimized version of popular diffusion models such as stable diffusion 3.5 and flux.
Additionally, the update to Nvidia RTX Remix, the platform that enables Modders Remaster Classic Games, launches today and adds an advanced path-tracing particle system that provides stunning visuals to infuse new life into classic titles.
Comfyui v3.57 boosts performance with RTX
Nvidia collaborated with Comfyui to improve the performance of its AI models by up to 40%. To take this into consideration, GPU generation upgrades usually only provide a performance improvement of 20-30%.
Developers interested in optimizing the performance and efficiency of app spreading models can read more about how NVIDIA accelerates these workloads in the developer forum.
The cutting edge AI model accelerated by RTX
The incredible model for creating AI content has been released in the past few weeks, all of which are now available in Comfyui.
WAN 2.2 is a new video model that offers incredible quality and control for video generation on PCs. This is the latest model of Wan AI. It is a creative AI platform that offers an impressive lineup of AI models, including image-to-text, text to video, images to video, speeches to video, and more. The GeForce RTX and NVIDIA RTX Pro GPUs are the only GPUs that can run WAN 2.2 14B models with COMFYUI without significantly delaying the output. See the example below created with a single prompt: “The robot is cracking an egg, but accidentally bumps it outside the bowl.”
Qwen-Image is the foundational model for Alibaba's new generation of images, achieving significant advances in complex text rendering and accurate image editing. It is excellent at rendering complex text, handles complex editing and maintains both semantic and visual accuracy of the generated images. This model runs 7x faster on the GeForce RTX 5090 vs. Apple M3 Ultra.
New Flux from Black Forest Labs. 1Krea [dev] The AI model is an open weight version of the Krea 1, trained to provide powerful performance and produce more realistic and diverse images that do not contain supersaturated textures. Black Forest Labs calls the model “speeed an opinion” because it provides a wide variety of diverse and visually interesting images. This model runs 8x faster on the GeForce RTX 5090 vs. Apple M3 Ultra.
Hunyuan3d 2.1 is a completely open source, production-ready 3D generation system that converts input images or text into high fidelity 3D assets enriched with physically-based rendering materials. The core components include 3.3 billion parameter models for shape generation and 200 million parameter models for texture analysis to quickly generate more realistic materials. It all runs faster on a Blackwell RTX GPU.
Get started with advanced visual generation techniques
Visual Generation AI is a powerful tool, but it can be difficult to get started, even by learning to use technical experts and more advanced technologies.
COMFYUI can easily get started with advanced workflows by keeping characters constant throughout different generations and providing templates or preset nodes that accomplish specific tasks, such as adjusting image light or loading tweaks. This makes it easy for even non-technical artists to use advanced AI workflows.
These are 10 important techniques to get started with Generating AI.
- Define start and end frames to guide video generation: How to upload a start and end frame and start and end video clips. WAN 2.2 allows you to generate smooth, animated transitions and fill in frames between to create coherent animations. Perfect for defining animation, scene shifts, or poses.
- Edit images in natural language:Uses Flux.1 [dev] A Kontext to edit a specific text section of an image.
- High-end images or videos: Acquire images or videos at low resolution by adding realistic and high frequency details, increasing resolution and quality of detail.
- Control aREA configuration: Assume that you have more control over image generation by controlling the placement and layout of visual elements within a particular area of the image.
- Restyle images: Use Flux Redux to create images in a variety of variations, while preserving core visual elements and details.
- Tap Image to 3D model: Create a high fidelity texture 3D model using multiple images of objects captured from different angles.
- Convert sound to video: Create video clips or animations directly from audio inputs such as audio, music, and environmental sounds.
- Control the video trajectory: Automatically guides the movement of objects, cameras, or scenes in the video.
- Edit images with Inpainting: Fill or modify missing or unnecessary parts of a digital image in a visually seamless and contextually consistent way.
- Expand the canvas with paint: Generate new image content to extend the boundaries of existing images or video footage, adding details to cropped sections or complete the scene.
Follow Comfyui on X for creative templates and workflow updates.
Expand the Comfyui zone
The COMFYUI plugin allows users to add generated AI workflows to existing applications. The Comfyui community has begun building plugins for some of the top popular creative applications.
Adobe Photoshop plugin complements Photoshop's native Firefly models by allowing users to run their own flows and select specialized models for specific tasks. Local inference also allows for unlimited generation filling with low latency.
Connect 2D and 3D workflows with the Blender plugin featured in the NVIDIA AI Blueprint of 3D Inductive Generation AI. Artists can use 3D scenes to control image generation, create textures in Comfyui, and apply 3D assets separately.
A similar Foundry Nuke plugin to Blender allows connections between 2D and 3D workflows, so users do not need to exchange Alt-Tabs between applications.
The Unreal Engine plugin allows COMFYUI nodes directly in the Unreal Engine user interface, and quickly create and refine scene textures using generative diffusion models. See the example below.
Running over-optimized models of Nvidia RTX GPUs in Comfyui
The best way to use NVIDIA RTX GPUS is to find it in the Tensort Library. It is a high-performance, deep learning inference engine designed to squeeze maximum speed out of the tensor core of an NVIDIA RTX GPU.
Nvidia is working with top AI labs to integrate Tensort into models, including Black Forest Labs models and Stability AI models. These models are quantized. This is a compressed version of the network that uses 50-70% less VRAM and offers inference up to twice as fast while maintaining similar quality.
The Tensortort-Optimized model can be run directly in Comfyui via the Tensort node currently supporting SDXL, SD3, SD3.5, and Flux.1-dev and Flux.1-schnell models. The node converts the AI model to the Tensortort-Optimized model and generates the Tensortort-Optimized engine for the user's GPU. This is a map of how to run that model with optimal efficiency for a particular hardware, providing a significant speedup.
However, quantization of the model requires a little more work. nvidia provides pre-configured files in a simple container called NIM microservices for users interested in running quantized and tensort-optimized models. Users can load these containers using Comfyui's NIM nodes and use quantized versions of models such as Flux.1-dev, Flux.1-Schnell, Flux.1 Kontext, SD3.5 Large, Microsoft Trellis, and more.
Remix Update adds a Path Trace Particle System
The new RTX remix update released today via the NVIDIA app adds an advanced particle system that allows modders to enhance the effects of traditional fire and smoke, allowing for more fantastic effects like video games Portal.
https://www.youtube.com/watch?v=4pt5t0mh9fk
With RTX Remix, legacy particles in classic games can be interpreted as pass traces, allowing you to cast realistic light to enhance the appearance of many scenes. But in the end, these particles were still over 20 years old, with no details, talent or fluid animations.
The new particles in RTX Remix have physically accurate properties and can interact with game lighting and other effects. This allows particles to collide and move accurately in response to wind and other forces, reflecting on the surface, casting shadows and assigning their own shadows.
Read our GeForce article for a complete breakdown of the new particle system.
Every week, RTX AI Garage The blog series features community-driven AI innovation and content for those looking to learn more about NVIDIA NIM microservices and AI blueprints, as well as buildings AI Agentcreative workflows, productivity apps, and more.
Connect to nvidia ai pc on Facebook, Instagram, Tiktok and x – And you will be notified by subscribing to RTX AI PC Newsletter. Join Nvidia's Discord Server Connect with community developers and AI enthusiasts to discuss what RTX AI is capable of.
Follow Nvidia Workstation LinkedIn and x.
look Let me know Regarding software product information.
