L2P: Unlocking Latent Potential for Pixel Generation

TL;DR

L2P introduces an efficient transfer framework that leverages pre-trained latent diffusion models to generate high-resolution pixel images with minimal training resources.

Contribution

The paper proposes L2P, a novel transfer paradigm that bypasses VAE training, enabling rapid, high-quality pixel generation from latent models using synthetic data and shallow training.

Findings

L2P achieves comparable performance to source LDMs on benchmark tasks.

L2P enables 4K ultra-high resolution image generation.

Training overhead is negligible compared to traditional methods.

Abstract

Pixel diffusion models have recently regained attention for visual generation. However, training advanced pixel-space models from scratch demands prohibitive computational and data resources. To address this, we propose the Latent-to-Pixel (L2P) transfer paradigm, an efficient framework that directly harnesses the rich knowledge of pre-trained LDMs to build powerful pixel-space models. Specifically, L2P discards the VAE in favor of large-patch tokenization and freezes the source LDM's intermediate layers, exclusively training shallow layers to learn the latent-to-pixel transformation. By utilizing LDM-generated synthetic images as the sole training corpus, L2P fits an already smooth data manifold, enabling rapid convergence with zero real-data collection. This strategy allows L2P to seamlessly migrate massive latent priors to the pixel space using only 8 GPUs. Furthermore, eliminating the VAE memory bottleneck unlocks native 4K ultra-high resolution generation. Extensive experiments across mainstream LDM architectures show that L2P incurs negligible training overhead, yet performs on par with the source LDM on DPG-Bench and reaches 93% performance on GenEval.