EditHero: A Benchmark for Long-Horizon Part-Level 3D Editing and Vibe Modeling
3D editing methods are usually tested on a single edit, yet an asset is built through a long sequence of revisions, each of which must implement the requested change while leaving everything else unchanged. We introduce EditHero, to our knowledge the first benchmark for long-horizon, part-level 3D editing, with natural-language instructions and target images for both geometry a
Reading in Plain English mode — toggle below to switch
3D editing methods are usually tested on a single edit, yet an asset is built through a long sequence of revisions, each of which must implement the requested change while leaving everything else unchanged. We introduce EditHero, to our knowledge the first benchmark for long-horizon, part-level 3D editing, with natural-language instructions and target images for both geometry a
This benchmark addresses a critical gap in 3D editing by focusing on long-horizon, part-level modifications, which is essential for developing robust and practical AI agents for content creation. It highlights the shift towards agentic methods that can iteratively inspect and modify 3D assets, moving beyond single-edit capabilities.