Adobe Research, in collaboration with UC San Diego, has developed EditAgent, a new artificial intelligence system designed to edit digital images similarly to a human professional using Adobe Photoshop. According to a report by The Decoder, the system departs from standard general-purpose AI editors that repeatedly apply the same tool. Instead, EditAgent selects specific specialized tools for individual editing tasks, while a standard language model evaluates and selects the optimal result using measured image data.
How Specialized Tools Change Image Editing
Most current consumer AI image editors rely on unified models that handle prompts broadly, often struggling with precise adjustments like local retouching or layered masking. By breaking the editing process down into discrete, specialized tool selection, EditAgent mimics a professional workflow where distinct instruments are chosen for specific problems. However, this precision comes with a steep computational trade-off. The research indicates that the system currently requires approximately ten times the amount of compute power compared to traditional approaches.
Unknowns and Future Performance
While the architectural approach shows promise, several practical details remain unaddressed. The initial announcement did not specify whether EditAgent will be integrated into commercial Adobe software such as Photoshop, nor did it provide a timeline for any potential public release. Additionally, future reports will need to clarify whether the tenfold increase in compute requirements can be reduced enough to make the tool viable for everyday consumer or enterprise use.
As research into tool-using AI agents continues, the balance between specialized execution and computational overhead will dictate whether systems like EditAgent move beyond the laboratory.
Lead image: Vito Goričan / Pexels
