GPT Image2's Core Team of 13 Revealed After 4-Month Redesign
Recently, GPT Image2 has taken social media by storm with its stunningly realistic image generation. As the project's popularity soared, the small, low-profile team behind it stepped into the limelight. Information reveals the core team comprises just 13 individuals, who managed to completely rewrite the underlying architecture in a mere four months. While research lead Chen Boyuan hasn't disclosed specific technical details, he describes this new model as "the GPT for the visual domain," signaling a major leap forward in general-purpose capabilities.
The team's key figure, Chen Boyuan , has a remarkable personal journey. During his PhD, he pioneered innovative approaches like "Diffusion Forcing" and contributed to instruction-tuning techniques later adopted by Google's Gemini 2.0. Interestingly, he didn't even know Python when he first joined a high school science camp. After moving to OpenAI, he not only led all training for the GPT image model but was also a core member of the Sora video generation team. In one demonstration, he highlighted the model's advanced language understanding by generating posters with perfectly rendered text in Chinese, Korean, and Bengali.

Beyond text rendering, GPT Image2 has achieved new heights in comprehending world knowledge and following complex instructions. This module, led by Dr. Jianfeng Wang from the University of Science and Technology of China, tackles a persistent issue in AI image generation—like older models always defaulting to drawing clocks at 10:10. The new model accurately interprets any specified time and complex spatial layouts. Dr. Wang notes the model is effectively closing the gap between a user's creative vision and the final generated result.
On the productivity front, Yuguang Yang from Zhejiang University's Zhuyuan College showcased the tool's ability to instantly convert lengthy research papers into high-precision PowerPoint presentations and infographics. This capability stems from the team's deep integration of multimodal understanding, Mixture of Experts (MoE) architecture, and long-context guidance technology.
From the original DALL-E to today's GPT Image2
Related article
Google Unveils Agentic AI for Gemini, Launching with Enterprise Clients
During a Google Cloud event on Thursday, the company unveiled plans to integrate Gemini AI into the agentic era. This launch introduces a unified agent capable of not only answering queries but also executing tasks on behalf of users, all within a si
Gecko Robotics partners with NVIDIA to integrate AI agent security and control
Gecko Robotics is leveraging NVIDIA’s latest agentic AI security layer within its inspection robot fleet. Source: Gecko RoboticsNVIDIA Corp. unveiled its Open Agent Safety Platform today, featuring the open-source OpenShell software and the Sentry re
Google Gemini Upgrades Floating Bubbles to Promote Lightweight AI Assistants
Starting September 6, Google initiated a phased rollout of the updated "minimized" quick action button featuring the Gemini interface for Android users. This update enables users to collapse the expansive Gemini overlay into a compact floating bubble
Related Special Topic Recommendations
Comments (0)
0/500
Recently,
The team's key figure,

Beyond text rendering,
On the productivity front, Yuguang Yang from Zhejiang University's Zhuyuan College showcased the tool's ability to instantly convert lengthy research papers into high-precision PowerPoint presentations and infographics. This capability stems from the team's deep integration of multimodal understanding, Mixture of Experts (MoE) architecture, and long-context guidance technology.
From the original DALL-E to today's
Google Unveils Agentic AI for Gemini, Launching with Enterprise Clients
During a Google Cloud event on Thursday, the company unveiled plans to integrate Gemini AI into the agentic era. This launch introduces a unified agent capable of not only answering queries but also executing tasks on behalf of users, all within a si
Gecko Robotics partners with NVIDIA to integrate AI agent security and control
Gecko Robotics is leveraging NVIDIA’s latest agentic AI security layer within its inspection robot fleet. Source: Gecko RoboticsNVIDIA Corp. unveiled its Open Agent Safety Platform today, featuring the open-source OpenShell software and the Sentry re
Google Gemini Upgrades Floating Bubbles to Promote Lightweight AI Assistants
Starting September 6, Google initiated a phased rollout of the updated "minimized" quick action button featuring the Gemini interface for Android users. This update enables users to collapse the expansive Gemini overlay into a compact floating bubble





Home






