NVDA 223.96 ▲2.27%GOOGL 354.30 ▼0.96%MSFT 499.99 ▲0.03%AMD 483.36 ▼1.21%INTC 101.65 ▲1.84%TSMC 420.04 ▲0.44%AMZN 274.48 ▲0.82%META 592.10 ▲0.37%AAPL 313.33 ▲0.29%PLTR 172.01 ▲10.32%
Markets at last close

Nvidia · Apps

NVIDIA Launches Blueprint for 3D-Guided Generative AI Image Creation

·2 min read

NVIDIA has introduced the AI Blueprint for 3D-guided generative AI for RTX PCs, providing creators and developers with powerful tools to control the composition of Artificial Intelligence–generated images. Traditional text-based prompts have simplified scene creation, but they struggle with nuanced aspects like camera angles and object placement, leaving users wanting greater creative oversight. NVIDIA´s blueprint addresses these limitations by integrating a 3D scene draft as a depth map, crafted in Blender, which guides the image generator—FLUX.1-dev from Black Forest Labs—in conjunction with user prompts to deliver customized results.

This depth map-driven technique offers advantages over pure text input, enabling users to intuitively manipulate every aspect of a scene, from object location to camera viewpoint, without requiring intricate 3D models or detailed textures. The system´s foundation leverages ComfyUI for chaining generative AI models and includes a Blender plug-in for seamless integration. The workflow also incorporates NVIDIA´s NIM microservice to optimize deployment and speed when running FLUX.1-dev on GeForce RTX GPUs, utilizing TensorRT and quantized formats such as FP4 and FP8 for substantial performance gains and reduced memory requirements. A GeForce RTX 4080 GPU or higher is recommended for optimal use.

NVIDIA´s blueprint is designed to lower barriers for both AI artists and developers. It comes as a prebuilt package with Blender, ComfyUI, essential plug-ins, deployment instructions, and all necessary nodes and microservices, ensuring both an easy start for newcomers and a flexible platform for experienced developers to extend. The solution benefits from the high-speed inference made possible by NVIDIA´s latest RTX and Blackwell architectures, halving model size requirements compared to previous standards. As part of a broader suite of over ten available NIM microservices targeting diverse AI tasks, this blueprint marks a significant advance in generative visual workflows and is available for immediate download, supporting real-time experimentation and customization on RTX-enabled PCs and workstations.

Originally reported by blogs.nvidia.comRead the source →
Related coverage
All Nvidia news →