UFO: unified omni-condition alignment evaluation for multimodal image generation, +15.25% human correlation

ustc-community · hf · 2026-09-18

UFO is the first unified framework for evaluating omni-condition alignment in multimodal image generation, addressing the flaw that embedding- and MLLM-based methods assess each modality condition in isolation. Its Atomized Chain-of-Evaluation paradigm decomposes alignment into fine-grained, disentangled Atomic Evaluation Units, classifies them by modality relevance, and verifies each with general or dedicated functional calls. UFO achieves the highest correlation with human preferences (+15.25% average), and the team releases UFO-Bench for holistic evaluation of subject-driven customization models.

Original post →

More from Multimodal

Multimodal channel →