Adobe researchers develop EditAgent AI for professional image editing

Researchers from Adobe Research and UC San Diego have introduced EditAgent, an AI agent designed to mimic professional image editing workflows. Unlike standard general-purpose editors that often fail complex tasks, EditAgent breaks down requests into steps and selects specific tools for tasks such as sharpening, rotating, or zooming.

In a study involving 600 image edits, a general-purpose editor failed to produce usable results in 42.7 percent of cases. The researchers found that generating multiple parallel variants from the original image and having a language model evaluate them was more effective than iterative refinement, which tends to accumulate errors. Professionals preferred the parallel approach in 78.5 percent of comparative tests.