# A Frozen Pixel-Space Diffusion Model Can Guide Itself with Its Own Samples

Source: [A Frozen Pixel-Space Diffusion Model Can Guide Itself with Its Own Samples](<https://www.sota2.com/research/paper/6a70205decc699bfe185c0ab>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-08-07T06:00:02.199Z

Authors: Zixuan Fu, Chong Wang, Lanqing Guo, Kailai Zhou, Jiahao Nie, Bihan Wen

Published: 2026-07-31T07:52:08.000Z

## Abstract

Pixel-space diffusion models aim to learn an end-to-end generator directly over raw pixels. This is challenging because a single model must capture both global structure and local texture in the same high-dimensional space. While recent work improves pixel diffusion through alternative prediction targets, training objectives, and architectures, these advances typically require training a new model from scratch. We show there is a cheaper, complementary strategy: \\textbf{a frozen, pretrained pixel diffusion model can guide itself}. Our key observation is that intermediate layers of a pretrained pixel diffusion transformer can be decoded into coarse predictions that capture the main low-frequency structure, while the final layers progressively refine local, high-frequency details. We therefore attach a lightweight prediction head to an intermediate layer, keep the backbone frozen, and use the discrepancy between the intermediate and final predictions as a self-guidance direction during sampling. To train this head, we further find that real images are not necessary. Instead, model-generated samples suffice and even outperform real images for training the head, especially in enhancing the high-frequency components that pixel diffusion tends to underfit. Across multiple pixel diffusion models on ImageNet, our \\textbf{Synthetic Self-Guidance (SSG)} consistently improves generation while adapter training requires less than 1$\\%$ of full-model training compute: it reduces FID by over 50$\\%$ across the evaluated JiT variants without classifier-free guidance (CFG) and further improves strong baselines with CFG, e.g., JiT-H/16 from 1.86 to 1.67 and PixelREPA-H/16 from 1.81 to 1.59. Our code is available at https://github.com/zfu006/SSG.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2607.29122>)

[PDF](<https://arxiv.org/pdf/2607.29122.pdf>)

[Implementation](<https://github.com/zfu006/SSG>)
