LongBanana benchmark tests long-context multi-reference image synthesis
Original title: LongBanana: An Expert-Verified Benchmark for Long-Context Multi-Reference Image Synthesis
LongBanana includes 90 tasks with 4,670 expert evaluation items for agent-driven image generation.
LongBanana is the first expert-verified benchmark for long-context multi-reference image synthesis. Tasks require retrieving evidence, binding entities, and composing coherent images from dispersed references. Agent harnesses improve performance, with ReAct-Plan-Exec-Reflect workflow strengthening grounding.
- 90 tasks organized by 3-by-3 taxonomy
- 32 reference images, 20 constraints, 50 evaluation items
- 4,670 atomic checklist items
- Graph-grounded evaluation for fidelity and binding
From the original article
1 min read at huggingface.co
We introduce LongBanana, to our knowledge the first expert-designed and expert-verified benchmark for long-context multi-reference image synthesis in the age of agent-driven image generation. The task requires retrieving dispersed visual evidence, rejecting distractors, binding entities and relations, and composing one coherent final image from multiple references.
By Haoxiang Cao, Yuxuan Zhang, Penghui Du · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.