Animating arbitrary rigged 3D assets remains labor-intensive, while existing generative approaches rarely provide asset compatibility, language-based control, iterative refinement, and preservation of production assets. We introduce GenRA, an end-to-end framework for generating and editing skeletal animations on user-provided assets. GenRA conditions a multimodal large language model on the rig hierarchy and bind pose, multiview renders, natural-language instructions, and optional motion demonstrations compactly represented through keyframe decimation. The model predicts sparse joint-transformation keyframes, which are subsequently converted into smooth animation curves prior to being authored into the original asset without modifying its geometry, skinning, or rig hierarchy. GenRA requires neither task-specific training nor retargeting and supports zero- and few-shot generation with localized, compositional, temporal, and quantitative refinements. Across 1,080 blind judgments involving 40 diverse assets, GenRA attains a 74.5% tie-aware preference rate over Puppeteer, is preferred in every evaluated category, and reduces median generation time by 83% and cost by 84%.
Project page: https://macpaw.github.io/GenRA/
@misc{yaroshevych-etal-2026-opensun3d-genra-3d-animation,
author = {Andrii Yaroshevych and Pavlo Kryven and Mariya Hirna and Maksym Shamrai},
title = {{GenRA}: End-to-end Framework for {LLM}-Driven Animation of Rigged {3D} Models},
month = {September},
year = {2026},
note = {\emph{Accepted to OpenSUN3D @ ECCV 2026.} \url{https://research.macpaw.com/publications/genra}},
}