Adobe and UC San Diego researchers introduce EditAgent, an AI agent for image editing
Overview
Adobe Research and UC San Diego researchers introduce EditAgent, an image-editing agent that generates several candidate edits using general and specialized tools, then picks the best one.
The Decoder reports that a standard language model makes the selection when given verification data such as close-ups, change maps and measured brightness and color statistics.
On a 600-edit benchmark, the agent using Qwen3.8-27B picked a usable result 81.3 percent of the time, compared with 74.8 percent for the fine-tuned EditScore-72B. The report says the agent needs about ten times the compute of a single editor.
Written by AI from the articles below · updated Nov 1, 10:40 AM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- The DecoderNewsAdobe researchers test an image editing agent that verifies its own edits
AIAdobe Research and UC San Diego built EditAgent, an image editing agent that generates parallel candidates with general and specialized tools. A standard language model picks the best result when given verification data such as close-ups, change maps and measured brightness and color statistics. On a 600-edit benchmark, Qwen3.8-27B picked a usable result 81.3 percent of the time, versus 74.8 percent for the fine-tuned EditScore-72B, but the agent needs about ten times the compute of a single editor.
Heat Trend
- Current comparable heat
- 5
- Peak comparable heat
- 10Nov 1, 11:00 AM ET
- Change in past 24 hours
- -50%
The trend compares only continuously observed participants, so its scope may be smaller than the current heat total. Hover or tap for hourly heat; use arrow keys with a keyboard.