STAGE: Subspace-Targeted Affine Generative Erasure for Text-to-3D Models

2026-10-01

AI Summary

This study addresses the challenge of concept erasure in native 3D generative models, where existing methods struggle to simultaneously preserve geometry and texture and remain incompatible with multi-stage architectures. To this end, we propose STAGE, a framework that introduces a novel stage-aware erasure mechanism, decoupling text-to-3D concept removal into distinct structural and appearance phases. By leveraging low-dimensional subspace projection and least-squares affine correction under relaxed orthogonality constraints, STAGE achieves training-free concept elimination. Experimental results demonstrate that STAGE attains a comprehensive score of 66.7 on TRELLIS, significantly outperforming the baseline (53.2). These findings indicate that our approach effectively balances targeted concept forgetting with the preservation of unrelated content, offering a robust solution for multi-stage 3D generation pipelines.

Abstract

Concept erasure suppresses a target concept while preserving behavior on unrelated inputs. Existing closed-form methods were designed for 2D image diffusion and assume a single generative pathway, so one edit must cover geometry and texture at once. Native 3D generators, which synthesize structured 3D representations directly rather than by lifting 2D samples, violate this assumption. We show that shape and object concepts must be erased in the structural stage of the pipeline and material concepts in the appearance stage. We therefore formulate erasure in native text-to-3D as a stage-aware editing problem and introduce STAGE, a training-free, closed-form framework. STAGE confines each edit to the low-dimensional subspace spanned by the differences between erase and anchor embeddings, and relaxes the norm-preserving (orthogonal) constraint of prior editors into a least-squares affine correction that maps target activations onto safe anchors subject to a penalty on the displacement of retained prompts. The correction applies to the structural stage, the appearance stage, or both. We find that the stage an edit must reach is determined by concept type. On TRELLIS, the standard open native 3D generator, across 15 shape, material, and object concepts, STAGE reaches 66.7 on a composite score that balances forgetting the target concept against preserving everything else, aggregating CLIP-based semantic and physical metrics, versus 53.2 for the strongest adapted baseline. Code: https://github.com/gmum/STAGE/ Project Page https://gmum.github.io/STAGE/

Problem

• Concept Erasure
• Text-to-3D
• Native 3D Generators
• Stage-aware Editing

Innovation

• Concept Erasure
• Text-to-3D
• Training-free
• Affine Correction
• Stage-aware Editing