Adobe is testing how well AI agents can use editing tools like a Photoshop professional would

| Source: THE DECODER

Tags: Adobe Research, UC San Diego, EditAgent, image editing, AI agents, Photoshop

Adobe Research and UC San Diego built EditAgent, an image-editing agent that picks specialized tools per sub-step and has a standard LLM judge results from measured image data, beating prior agents at about ten times the compute.

Details

EditAgent breaks an image editing task into sub-steps and pairs a general-purpose editor with specialized tools for jobs like sharpening or zooming, instead of retrying one editor repeatedly. It generates image variants in parallel, and a standard language model evaluates them using measured statistics and close-up crops. With that data, the general model outperforms evaluation models trained specifically for the task. The system produces more accurate edits than previous AI agents but needs about ten times the compute. The researchers note the verified outputs can be used to train faster editing models, which is the more practical path for production tools.