Researchers have introduced TECCI, a new benchmark designed to rigorously test text-guided image editing models. TECCI comprises 7,550 image-edit instruction pairs, intentionally curated to expose weaknesses in current AI editing capabilities, particularly with challenging edits involving position, motion, and creativity. Human evaluations of leading models on TECCI revealed that none achieved over a 22% success rate, with Nano Banana Pro showing the best performance, though all models struggled more with minimal edits and visual quality than with instruction following. AI
IMPACT TECCI benchmark highlights significant limitations in current AI image editing, particularly for complex instructions, indicating a need for improved instruction following and visual fidelity.
RANK_REASON The cluster describes a new academic paper introducing a benchmark for evaluating AI models. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →