EditAgent uses tools and verification to improve AI image editing
Adobe Research and UC San Diego introduced EditAgent, an AI system that breaks image-editing requests into dependent steps and compares a general-purpose editor with specialized tools such as sharpening, rotation, and zooming. A standard language model selects among parallel candidates using close-up crops, change maps, and measured brightness and color statistics; on a 600-edit benchmark, all seven tested language models with verification data outperformed the fine-tuned EditScore-72B judge. The researchers also reported gains over prior agents and generated about 26,000 verified training examples, but said EditAgent requires roughly ten times the compute of a single editor.
The research suggests that structured verification records can make general language models more effective image-edit judges than models specifically trained for evaluation, while also supporting faster downstream editors through verified training data.
Adobe is testing how well AI agents can use editing tools like a Photoshop professional would
Jonathan Kemper View the LinkedIn Profile of Jonathan Kemper
Nov 1, 2026

Nano Banana Pro prompted by THE DECODER
Key Points
- Adobe Research and UC San Diego built EditAgent, an AI system that breaks image editing tasks into sub-steps and pairs a general-purpose editor with specialized tools for things like sharpening or zooming.
- A standard language model evaluates image variants generated in parallel, using measured statistics and close-up crops. With this data, it outperforms evaluation models that were specifically trained for the job.
- EditAgent produces more accurate results than previous AI agents but needs about ten times the compute. The verified output can then be used to train faster editing models.
#EditAgent#Adobe Research#UC San Diego#Qwen-Image-Edit-2511#Flux-2-Klein-9B#EditScore-72B#Qwen3.8-27B#Seedream 4.0#GPT-Image-1.5