EditAgent AI agent outperforms finetuned models in image editing
First merged summary.
EditAgent pairs general editors with specialized tools and uses parallel listwise judging to outperform finetuned reward models on 600-edit benchmarks.
Adobe researchers built EditAgent, an agentic AI system for image editing. It routes each step to general-purpose editors like Flux-2-Klein-9B and specialized methods while attaching verification records to every candidate. Off-the-shelf vision-language judges with these records outperform finetuned reward models, and professionals preferred parallel variants in 78.5% of decided comparisons. The system beat seven other models including EditScore-72B on 600 test edits and improved two open-weight editors on three benchmarks while generating 26,000 verified training examples.
- General-purpose editor fails 42.7% of 600 test edits
- Professionals prefer parallel listwise judging in 78.5% of comparisons
- EditAgent beats seven models including EditScore-72B
- Produces 26,000 verified training examples
- Improves two open-weight editors on three benchmarks
- Qwen3.8-27B scores 81.3% usable picks with verification data
Timeline
dated events in the reporting · oldest first- Oct 28, 2026Hugging Face daily papers introduces EditAgent for agentic image editing · Hugging Face daily papers
- Nov 1, 2026The Decoder reports Adobe testing EditAgent with 600-edit benchmarks · The Decoder
What each outlet adds
beyond the shared facts- Hugging Face daily papersoutperforms finetuned reward models with verification records and serves as data engine for training
- The DecoderAdobe built EditAgent for Photoshop-like tools and beats seven models including EditScore-72B
- Vendors
- Adobe
- Organizations
- Adobe ResearchUC San Diego
- AI models
- EditAgentQwen3.8-27B
Coverage
oldest first · one row per article · UTCEditAgent expands tools and verification for agentic image editing
Experts preferred parallel listwise judging over sequential refinement in 78.5% of decided comparisons.
Adobe develops EditAgent for professional image editing
EditAgent generates parallel variants and uses verification data to outperform fine-tuned models on 600-edit benchmarks.