Manuscript received August 7, 2024; revised August 14, 2024; accepted August 27, 2024; published November 18, 2024.

Abstract— Image Style transfer is a neural network algorithm that copies the style of an existing image into another image while preserving the image’s content. There have been various approaches on style transfer in an effort to speed up the process or provide more appealing results, one of which is the usage of style attentional networks. Attention is an algorithm that scores different parts of an image based on their importance in the overall image, attention helps neural networks distinguish important parts of an image. We use attention to identify the parts of an image that represent image style to apply an overall style rather than a mask and to conserve parts of the content that are crucial to its identity (a visible object, a focused subject, etc.). Aflutter Craft enhances an existing algorithm that uses attention for style transfer. Results show that our algorithm uniformly applies important parts of the style while simultaneously preserving the subject of the content image. Results from Aflutter Craft are chosen to be the most visually appealing according to 38.4% survey participants when compared to 4 other implementations. In addition, this paper introduces a cross platform application with a general Application Programming Interface (API) capable of performing style transfer from anywhere.

Keywords—Application Programming Interface (API), Art, convolutional neural network, deep learning, flask, neural network, PyTorch, self-attention, style transfer, transfer learning, transformer

Cite: Rawad Abdulghafor, "Aflutter Craft: Neural Art Transfer Platform," Journal of Image and Graphics, Vol. 12, No. 4, pp. 382-395, 2024.

Copyright © 2024 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC BY-NC-ND 4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.