Generating Images with a 2025 Android (www.duration.ai)

In an ambitious tech experiment, developers have successfully generated images using the PrismML's Bonsai Image model on the Samsung Galaxy S25+, highlighting significant strides in mobile AI capabilities. This project, which parallels a previous successful implementation on an iPhone, faced considerable challenges due to the less mature Android machine-learning ecosystem. While iOS offers robust tools like Core ML, Android's options, specifically LiteRT and NNAPI, led to complications regarding performance optimization across diverse hardware configurations. The final implementation on the S25+ utilized the phone's Hexagon NPU to achieve reasonable generation times, completing 512×512 images in about two minutes, despite encountering issues such as thermal throttling and precision limitations due to hardware constraints. This achievement is noteworthy for the AI/ML community as it demonstrates the practical application of advanced image generation models on consumer-grade mobile devices, paving the way for more accessible AI tools. The findings not only highlight the performance gap between iOS and Android environments but also underscore the potential for future enhancements in mobile AI functionalities. With the project serving as a foundation for further exploration, it invites other developers to refine the process and tackle the obstacles encountered, thus fostering innovation in the Android ecosystem.