Memory-Efficient Personalization of Text-to-Image Diffusion Models via Selective Optimization Strategies

TL;DR

This paper introduces a memory-efficient framework for personalizing text-to-image diffusion models on edge devices by adaptively combining backpropagation and zeroth-order optimization, guided by diffusion timesteps.

Contribution

It proposes a selective optimization strategy with a timestep-aware probabilistic function to improve personalization efficiency and quality while reducing memory usage.

Findings

Achieves high-quality personalization with reduced memory footprint.

Effectively balances adaptation speed and structural consistency.

Enables scalable on-device diffusion model customization.

Abstract

Memory-efficient personalization is critical for adapting text-to-image diffusion models while preserving user privacy and operating within the limited computational resources of edge devices. To this end, we propose a selective optimization framework that adaptively chooses between backpropagation on low-resolution images (BP-low) and zeroth-order optimization on high-resolution images (ZO-high), guided by the characteristics of the diffusion process. As observed in our experiments, BP-low efficiently adapts the model to target-specific features, but suffers from structural distortions due to resolution mismatch. Conversely, ZO-high refines high-resolution details with minimal memory overhead but faces slow convergence when applied without prior adaptation. By complementing both methods, our framework leverages BP-low for effective personalization while using ZO-high to maintain structural consistency, achieving memory-efficient and high-quality fine-tuning. To maximize the efficacy of both BP-low and ZO-high, we introduce a timestep-aware probabilistic function that dynamically selects the appropriate optimization strategy based on diffusion timesteps. This function mitigates the overfitting from BP-low at high timesteps, where structural information is critical, while ensuring ZO-high is applied more effectively as training progresses. Experimental results demonstrate that our method achieves competitive performance while significantly reducing memory consumption, enabling scalable, high-quality on-device personalization without increasing inference latency.