Mobile anime character generation using artificial intelligence and interactive creative pathways

This study examines the generation of anime characters using artificial intelligence on mobile devices, emphasizing interactive creative pathways. In this context, users iteratively guide content creation through natural language prompts, detailed attribute controls, and multistep refinement processes. We introduce MobileAnimeGen, a lightweight, mobile-aware generative framework based on a compressed latent diffusion architecture, designed to function entirely within the computational limits of modern smartphones. The framework incorporates a three-stage optimization pipeline, including structured pruning, mixed-precision quantization (W4A8), and progressive knowledge distillation, resulting in a 3.8× reduction in model size and a 4.6× increase in inference speed, with minimal loss in visual quality. The training and evaluation processes utilize publicly available datasets, such as a curated subset derived from Danbooru and the Manga109 corpus, ensuring complete reproducibility. We conceptualize interactive creative pathways as structured sequences of user actions and system responses, introducing quantitative metrics for pathway depth, exploration breadth, and co-creative effectiveness. A comprehensive user study with 48 participants across three levels of expertise indicates that the proposed adaptive interaction design significantly enhances perceived creative control (mean Likert score 4.3/5) and user satisfaction (4.4/5) compared to both cloud-based baselines and non-interactive on-device alternatives. The experimental findings reveal favorable trade-offs between image quality (FID 41.2 on the Danbooru test split), controllability, latency (1.6 s on a mid-range device), and energy efficiency, establishing a reproducible benchmark for mobile anime generation with a human-centered analysis of how mobile constraints influence co-creative AI systems.