Some developers have developed StableDesign project, which aims to create a new generation ofInterior DesignProvide datasets and training methods. This is a SD program for interior design, upload a picture of an empty room, enter text prompts, and you can generate a decorative rendering as required.
They first downloaded the listing data via Audemars Piguet and extended the bnb-dataset code with scripts to extract image metadata from it. They then applied feature extraction and data preparation tools to create a dataset suitable for interior design in preparation for subsequent model training.
After the data was prepared, the research team trained it using ControlNet and the model.The ControlNet model enables control over the interior design generation process by segmenting the image and depth estimating the image. Meanwhile, the model enables users to generate corresponding interior design images from textual descriptions through text-to-image conversion. These training methods combine image processing and natural language processing techniques and provide new ideas and methods for the development of the field of generative interior design.
The launch of the StableDesign project has facilitated the development and innovation of generative interior design techniques by providing researchers and designers with a new dataset and training framework. However, the results generated by StableDesign can't be trusted, for example, sometimes load-bearing walls are lost.
Online experience:https://huggingface.co/spaces/MykolaL/StableDesignGithub.