Search academic texts

Information × Registration Number 2124U004696, Article popup.category Препринт Title popup.author Maksymiuk Viktoriia popup.publication 01-01-2024 popup.source_user Український католицький університет popup.source https://hdl.handle.net/20.500.14570/4843 popup.publisher Description The rapid development of generative artificial intelligence (AI) has opened up a variety of opportunities to revolutionize the realm of creative content generation across diverse industries. Currently, content creation is a labor-intensive and time- consuming process, yet it remains in high demand, especially in the gaming, an- imation, and commercial industries. This work delves into integrating advanced generative AI models, particularly the Stable Diffusion (SD) model, with Blender, a leading computer graphics software, to automate the creation of 2.5D content. The primary objective is to develop an innovative automated pipeline capable of gen- erating realistic 2.5D scenes from textual prompts seamlessly while allowing for re- alistic depth-guided placement of user-provided 3D objects within the scene. The presented approach leverages the capabilities of SD for high-quality scene image generation and Marigold model for detailed depth map estimation. Integration with Blender software via Python API simplifies the content generation process to a few steps, making it accessible even to users with limited experience in 3D modeling. The results demonstrate that the proposed solution efficiently generates high- quality 2.5D scenes, providing realistic placement of objects within the generated environment, which is facilitated by the key component of accurate estimation of surface normal vectors. Additionally, the comprehensive setup incorporates light- ing and rendering configurations to deliver visually appealing and realistic content. To accommodate users with varying computational resources, two pipeline versions have been developed: a GPU-accelerated version that executes the entire workflow locally and a CPU-based version that leverages external resources for specific tasks. Both versions exhibit satisfactory time efficiency, offering a streamlined approach to content creation with minimal user intervention. By automating 2.5D content creation with depth-guided object placement, this work advances the field of au- tomated content creation, showcasing the potential of generative AI in simplifying workflows that traditionally demand extensive manual effort in 3D scene modeling and composition. popup.nrat_date 2025-05-09 Close