Year: 2024
Type: article
Abstract: Text-to-image creation systems are developed as part of image processing, allowing users to produce visual representations from written descriptions. This seeks to construct a system with a variety of essential capabilities, like making several photos with a single question, adding negative prompts to omit specified sections, and changing already-existing images with textual prompts. In addition to libraries like accelerate, transformers, ftfy, bits and bytes, gradio, natsort, safetensors, and xformers for effective model training, data processing, and user interface development, the approach leverages robust deep learning frameworks and libraries like PyTorch, TorchVision, TorchAudio, and diffusers for text-to-image generation. The system's user-friendly design features a web interface with stream lighting that enables users to submit unique photos for modification, enter textual prompts, and establish negative prompts. With the support of huge datasets utilized for training, the underlying models are able to produce meaningful images in response to provided instructions. Applications of these systems in creative design and human-computer interaction have increased interest in them.
Source: Library Progress (International)
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W7218268172