Conference Papers Year : 2022

Converting Nebulous Ideas to Reality – A Deep Learning Tool for Conditional Synthesis of Character Designs

Lilian Guo
  • Function : Author
  • PersonId : 1334349
Anand Bhojan
  • Function : Author
  • PersonId : 995220

Abstract

Characters are essential elements of games and are critical to their success. At the same time, designing good characters can be time and labor intensive, especially when developing games with thousands of characters. In such cases, procedural generation may be used to expedite the process. However, characters generated by traditional procedural generation techniques often rely on a limited pool of premade assets and may lack novelty. This work explores deep learning for the conditional generation of creative character designs with artist input. It proposes a framework which receives artists’ inputs in the form of blurred character silhouettes and converts these into high resolution character designs using Generative Adversarial Networks. In addition, the paper presents a demo Graphical User Interface and user study evaluating the tool’s effectiveness.
Fichier principal
Vignette du fichier
527508_1_En_7_Chapter.pdf (634.79 Ko) Télécharger le fichier
Origin Files produced by the author(s)

Dates and versions

hal-04388149 , version 1 (11-01-2024)

Licence

Identifiers

Cite

Lilian Guo, Anand Bhojan. Converting Nebulous Ideas to Reality – A Deep Learning Tool for Conditional Synthesis of Character Designs. 6th International Conference on Computer, Communication, and Signal Processing (ICCCSP), Feb 2022, Chennai, India. pp.75-89, ⟨10.1007/978-3-031-11633-9_7⟩. ⟨hal-04388149⟩
31 View
2 Download

Altmetric

Share

More