TIMI: Training-Free Image-to-3D Multi-Instance Generation with Spatial Fidelity

Xiao Cai, Pengpeng Zeng, Ji Zhang, Heng Tao Shen, Jingkuan Song, Lianli Gao
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:10789-10800, 2026.

Abstract

Precise spatial fidelity in Image-to-3D multi-instance generation is critical for downstream real-world applications. Recent work attempts to address this by fine-tuning pre-trained Image-to-3D (I23D) models on multi-instance datasets, which incurs substantial training overhead and struggles to guarantee spatial fidelity. In fact, we observe that pre-trained I23D models already possess meaningful spatial priors, which remain underutilized as evidenced by instance entanglement issues. Motivated by this, we propose TIMI, a novel Training-free framework for Image-to-3D Multi-Instance generation that achieves high spatial fidelity. Specifically, we first introduce an Instance-aware Separation Guidance (ISG) module, which facilitates instance disentanglement during the early denoising stage. Next, to stabilize the guidance introduced by ISG, we devise a Spatial-stabilized Geometry-adaptive Update (SGU) module that promotes the preservation of the geometric characteristics of instances while maintaining their relative relationships. Extensive experiments demonstrate that our method yields better performance in terms of both global layout and distinct local instances compared to existing multi-instance methods, without requiring additional training and with faster inference speed.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-cai26n, title = {{TIMI}: Training-Free Image-to-3{D} Multi-Instance Generation with Spatial Fidelity}, author = {Cai, Xiao and Zeng, Pengpeng and Zhang, Ji and Shen, Heng Tao and Song, Jingkuan and Gao, Lianli}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {10789--10800}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/cai26n/cai26n.pdf}, url = {https://proceedings.mlr.press/v306/cai26n.html}, abstract = {Precise spatial fidelity in Image-to-3D multi-instance generation is critical for downstream real-world applications. Recent work attempts to address this by fine-tuning pre-trained Image-to-3D (I23D) models on multi-instance datasets, which incurs substantial training overhead and struggles to guarantee spatial fidelity. In fact, we observe that pre-trained I23D models already possess meaningful spatial priors, which remain underutilized as evidenced by instance entanglement issues. Motivated by this, we propose TIMI, a novel Training-free framework for Image-to-3D Multi-Instance generation that achieves high spatial fidelity. Specifically, we first introduce an Instance-aware Separation Guidance (ISG) module, which facilitates instance disentanglement during the early denoising stage. Next, to stabilize the guidance introduced by ISG, we devise a Spatial-stabilized Geometry-adaptive Update (SGU) module that promotes the preservation of the geometric characteristics of instances while maintaining their relative relationships. Extensive experiments demonstrate that our method yields better performance in terms of both global layout and distinct local instances compared to existing multi-instance methods, without requiring additional training and with faster inference speed.} }
Endnote
%0 Conference Paper %T TIMI: Training-Free Image-to-3D Multi-Instance Generation with Spatial Fidelity %A Xiao Cai %A Pengpeng Zeng %A Ji Zhang %A Heng Tao Shen %A Jingkuan Song %A Lianli Gao %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-cai26n %I PMLR %P 10789--10800 %U https://proceedings.mlr.press/v306/cai26n.html %V 306 %X Precise spatial fidelity in Image-to-3D multi-instance generation is critical for downstream real-world applications. Recent work attempts to address this by fine-tuning pre-trained Image-to-3D (I23D) models on multi-instance datasets, which incurs substantial training overhead and struggles to guarantee spatial fidelity. In fact, we observe that pre-trained I23D models already possess meaningful spatial priors, which remain underutilized as evidenced by instance entanglement issues. Motivated by this, we propose TIMI, a novel Training-free framework for Image-to-3D Multi-Instance generation that achieves high spatial fidelity. Specifically, we first introduce an Instance-aware Separation Guidance (ISG) module, which facilitates instance disentanglement during the early denoising stage. Next, to stabilize the guidance introduced by ISG, we devise a Spatial-stabilized Geometry-adaptive Update (SGU) module that promotes the preservation of the geometric characteristics of instances while maintaining their relative relationships. Extensive experiments demonstrate that our method yields better performance in terms of both global layout and distinct local instances compared to existing multi-instance methods, without requiring additional training and with faster inference speed.
APA
Cai, X., Zeng, P., Zhang, J., Shen, H.T., Song, J. & Gao, L.. (2026). TIMI: Training-Free Image-to-3D Multi-Instance Generation with Spatial Fidelity. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:10789-10800 Available from https://proceedings.mlr.press/v306/cai26n.html.

Related Material