The Digital Press

A new study analyzes 83,000 real-world image editing requests to evaluate how well generative AI tools perform everyday photo edits compared to humans. It finds that current AI can fulfill about one-third of these requests, often struggling with precise edits and identity preservation.

Why it matters: Understanding AI’s strengths and weaknesses guides improvements in generative AI for practical image editing applications.

Stunning stat: Only around 33% of everyday image editing requests can be successfully completed by the best AI editors like GPT-4o and Gemini-2.0-Flash.

The big picture: AI performs better on creative, open-ended edits but struggles with precise, low-creativity tasks and maintaining subject identity.

Commenters say: Many note the gap between AI and human editors in precision, while some highlight AI’s potential in stylized, less exact edits.