AI TrackerFrontier desk · AEST
Back to news
normalaitech

MoI-Bench: A Benchmark for Image Generation and Editing on Mobile Devices

arXiv cs.AI · 28 Oct 2026, 03:31 AEST

MoI-Bench is a new benchmark designed for image generation and editing on mobile devices, addressing user needs and resource constraints. It includes 64 subtasks and 1150 samples to guide the development of on-device models compared to server-side models.

Key facts

Read aloud

MoI-Bench is a new benchmark designed for image generation and editing on mobile devices, addressing user needs and resource constraints. It includes sixty four subtasks and one thousand one hundred fifty samples to guide the development of on-device models compared to server-side models. Developing on-device image models requires considering practical user needs, task types and device resource constraints. We introduce MoI-Bench, a benchmark grounded in mobile user needs, comprising sixty four subtasks and one thousand one hundred fifty samples. We further propose MoI-Eval, which introduces task-specific criteria covering instruction following, visual quality and detail preservation. Results show that this decoupled Operation x Target design provides finer-grained guidance for improving compact models. Further progress requires improving editing quality, particularly identity preservation and text rendering, while broadening task coverage.

Sources in this cluster