From a Single Design to Multiple Variations: AI-Guided in-App Image Generation
Gali Hod, Dvir Ben Or, Eyal Regev · 2025
In mobile game design, 2D artists and designers often spend a significant amount of time on repetitive tasks that detract from their creative process. One major pain point is the manual resizing and modification of pop-up images for various screen resolutions, a process that can consume up to 50% of the designers' time after the initial creative work is complete. This Sisyphean task, involving extensive manual edits across multiple Photoshop files, inhibits productivity and limits time spent on more valuable creative work. Despite recent advances in generative AI, no existing solution fully addresses this specific challenge. In this work, we propose and implement an AI-based system tailored for artists, which takes a Photoshop file organized in layers (.psd/.psb) of an in-app image at any resolution and generates layered variations at different resolutions. These variations retain the essence of the original content while being intentionally distinct. As the output is provided as a layered Photoshop file, artists have greater freedom for further creative refinement. The Code is Available at: https://github.com/PlaytikaOSS/layout-design-variations