Skip to content
TopicTracker
From HackerNewsView original
TranslationTranslation

Generating Images with a 2025 Android

The article explores the state of AI image generation on Android devices in 2025, covering native on-device models, cloud-based tools, and practical workflows for creating images directly from a smartphone.

Background

- Duration AI is a startup (backed by a16z) building on-device AI models. This post shows images generated entirely on a 2025 Android phone, with no cloud processing. - The article demonstrates that modern mobile chipsets (e.g., Qualcomm Snapdragon 8 Gen 4, MediaTek Dimensity 9400) can now run diffusion models (like Stable Diffusion) locally at usable speeds — a task that required a powerful desktop GPU just a year or two ago. - This matters because on-device AI preserves privacy (data never leaves the phone), works offline, and avoids cloud compute costs. It signals a shift from "AI in the cloud" to "AI in your pocket." - The images shown are not curated cherry-picks; the post discusses generation latency (seconds per image), resolution limits, and the trade-offs of running large models on limited memory.

Related stories