Google DeepMind Introduces CO2Jump for Joint Text and Image Generation

Researchers from Google DeepMind have developed CO2Jump, a training-free sampler that allows text and images to dynamically correct each other during the denoising process.

Compiled by Sergey KostenchukPublished 2026-07-20Updated 2026-07-21

2026-07-20 Research Google

🖼 Google DeepMind Introduces CO2Jump for Synchronous Text and Image Generation

Researchers have developed CO2Jump, a training-free sampler based on the SC-CMJP mechanism. It allows text and images to dynamically correct each other during the denoising process, fixing errors on the fly.

🌍 This solves the "uncoupling failure" problem without the cost of retraining models.

👤 The AI gains a better understanding of the relationship between the description and the visual content.

Source 1: https://arxiv.org/abs/2607.13188 Source 2: https://coupled-jump.github.io/

Sources