The Generative Interior Design Challenge: How Did the Challenge Winners Transform Empty Rooms with AI?

The Generative Interior Design Challenge is the latest interdisciplinary competition hosted on AIcrowd that uses Generative AI to enhance Interior Design. Participants were tasked with using AI to transform empty rooms into furnished spaces based on text descriptions. The goal was to develop a model capable of understanding a text prompt and interpreting an image of an empty room to generate a new image that includes furnishings and design prompts.

The Objective

This challenge required participants to blend their technical skills and creativity to produce aesthetically pleasing and functional designs. The goal was to change how designers, architects, and homeowners envision and plan interior spaces, making the process more efficient and accessible.

Participants used LLMs and Stable Diffusion Models to create visually appealing interiors based on text prompts and a reference image of an empty room to create visually appealing interiors based on text prompts. Submissions were evaluated on room layout, realism, functionality, and consistency with the text prompt. Generated images were also manually assessed by a panel of trained experts to ensure that the designs adhered to the original room’s structure and met standards of realism and practicality.

The amalgamation of art and technology in this challenge provided a chance to explore the impact of generative networks in advancing the interior design field and opening new avenues for client-designer interactions.

MEET THE WINNERS

1st Place: Team decem

• Prize: $7,000
• Members: Yi Zhang, Chenyue Wang, Xiaoya Liu

Background: Yi Zhang graduated with a bachelor’s degree and currently works at Sichuan Timesnew Technology Co., Ltd. Along with Chenyue Wang, and Xiaoya Liu, students of Peking University, the team blends experience and enthusiasm. They joined this challenge to improve their understanding and practical skills in diffusion and ControlNet.

Approach: The team began by learning and understanding the concepts of diffusion and ControlNet. Applying their theoretical knowledge to real-world scenarios through hands-on projects, they embraced setbacks and saw them as opportunities for growth. They prepared for the challenge with rigour and practice, ultimately topping the leaderboard.

2nd Place: Team StableDesign

• Prize: $5,000
• Members: Mykola Lavreniuk, Bartosz Ludwiczuk

Background: Mykola, a PhD in Applied Mathematics and Senior ML/CV Engineer at the Space Research Institute, holds expertise in visual analysis, detection, segmentation, and monocular depth estimation. He has developed state-of-the-art models and has experience in multimodal techniques, generative AI, and satellite image analysis. His teammate, Bartosz Ludwiczuk, is the Computer Vision Tech Lead at SoftServe. He specialises in representation learning techniques in the image domain, with experience in face recognition, image retrieval, and gait recognition. Both Mykola and Bartosz are regular participants in Machine Learning competitions, consistently achieving top ranks in AIcrowd competitions.

Approach: Team StableDesign started by developing a dataset, sourcing approximately 300,000 images from Airbnb and processing them to extract features for interior and outdoor scenes. They filtered out indoor images, generated segmentation masks and depth maps for each, and created empty room images using an inpainting pipeline. They used the llava1.5 model to generate image descriptions and trained two custom Control Nets based on depth and segmentation-conditioned images. During inference, they used an inpainting pipeline based on the Realistic_Vision v5.1 model, along with the two custom Control Nets and the IP Adapter. Their full code is available here, and their final best model can be accessed here.

3rd Place: Team XenonStack

• Prize: $3,000
• Members: Akash Pandey, Navdeep Singh Gill, Dr Jagreet Kaur

Background: Akash is an AI Solution Architect at Xenonstack, specialising in providing Generative AI solutions and overseeing AI operations. He is passionate about advancing technology through experimental projects. Xenonstack is a Technology Consulting and Solutions company focusing on platform engineering, data intelligence, and generative and edge AI. The team is led by Navdeep Singh Gill and Dr Jagreet Kaur, who have over 20 years of experience in the field and are recognised as LinkedIn’s Top AI and Data Voices.

Approach: Team XenonStack began by analysing the problem statement and baseline solution and identifying enhancement opportunities. They implemented a pipeline with a base model, exploring various prompt strategies. Their top improvements included integrating dual ControlNet models, advanced segmentation, masking models, and inpainting models to refine detail and realism. Each modification was tested and improved iteratively. Despite challenges, they continuously adapted by enhancing their strategies to meet the design criteria better, leading to their eventual victory.

Thank you to all participants, and congratulations to our winners! Their innovative approaches and hard work helped redefine the impact of AI on interior design.