Adobe and UC San Diego propose EditAgent, an image-editing agent
What happened
On November 1, The Decoder reported that Adobe Research and UC San Diego proposed EditAgent, an image-editing agent that calls general editors and specialized tools in parallel and uses quantitative verification data to pick among edit results. Across 600 edit tests, Qwen3.8-27B with verification data selected a usable result 81.3% of the time, above EditScore-72B's 74.8%. The report says EditAgent improved open-source editors on three benchmarks, but needs about ten times the compute of a single editor, and its performance depends on the tools it is given.
Written by AI from the coverage · updated 2 hours ago
Coverage
Follow the reports to see the story from different sides.
- The DecoderAdobe is testing how well AI agents can use editing tools like a Photoshop professional would
Adobe Research与UC San Diego提出EditAgent，通过并行调用通用编辑器和专用工具、结合量化核验数据选择图像编辑结果。在600项编辑测试中，获得核验数据的Qwen3.8-27B选中可用结果的比例为81.3%，高于EditScore-72B的74.8%。EditAgent在三个基准上改善了开源编辑器的表现，但所需计算量约为单个编辑器的十倍，性能也取决于接入的工具。
Heat over time
Not enough continuous observations to draw a trend yet.