# PEEK Aims to Reduce Diffusion Image Errors During Sampling

> Researchers propose a training-free sampler plugin that they say improves image-generation scores across eight diffusion models, with a small calibration cost.

Oossa · 2026-11-05 · https://oossa.com/en/peek-aims-to-reduce-diffusion-image-errors-during-sampling

Researchers introduced PEEK, a training-free plugin for diffusion image generators. It makes an extra denoiser evaluation during selected sampling steps, then uses that look-ahead prediction to adjust the next step and reduce exposure bias—the mismatch between training inputs and the imperfect inputs a model encounters while generating.

The authors report improvements over baselines matched for denoiser evaluations across eight pixel- and latent-space models, including text-guided setups. They report FID improvements of up to 49.3%; calibrating the plugin for a model took at most 2.2 minutes in their experiments.

## The facts

- The paper was accepted to NeurIPS 2026.
- The authors tested PEEK on eight diffusion models and report FID improvements up to 49.3%.

## Why it matters

For teams researching diffusion image generators, PEEK offers a sampler-side method to test without retraining, though the paper does not establish how it performs in deployed products.

## Sources & references

1. [PEEK: One-Step-Look-Ahead Exposure Bias Correction for Diffusion Sampling](https://arxiv.org/abs/2611.05407) – arXiv cs.CV, 2026-11-05

Last updated: 2026-11-05