Jiahui Yu
Jiahui Yu is an AI researcher and lab leader whose work spans computer vision, image generation, and multimodal systems. His early papers introduced object-detection and image-inpainting methods; later work included VQGAN and the Parti text-to-image model. At Google DeepMind, he co-led Gemini’s multimodal work; at OpenAI, he coauthored the GPT-4o System Card and led multimodal training for o3 and o4-mini. Meta hired him in 2025 for its Superintelligence group, where he led work on Muse Spark; as of October 2026, he is starting a new company.
- Now at
- Google DeepMind
- Works
- 183
- Citations
- 14,605
- h-index
- 38
Publication numbers from OpenAlex, as of 15 October 2026.
Timeline
-
2016–2018
Object detection and image inpainting
In 2016, he coauthors UnitBox, which uses an intersection-over-union loss to predict object bounding boxes. In 2018, he coauthors a gated-convolution system for completing images with free-form masks.
-
2021–2022
Image models with discrete tokens
In 2021, he coauthors work on improved Vision-Transformer-based VQGAN, which encodes images as discrete tokens for autoregressive modeling. In 2022, he coauthors Parti, an autoregressive text-to-image model that generates high-fidelity images and supports content-rich synthesis involving complex compositions and world knowledge.
-
2023
Co-leading Gemini multimodal work
At Google DeepMind, he co-leads Gemini’s multimodal pillar in 2023. The Gemini report describes image, audio, video, and text understanding across the model family.
-
2024
Contributing to GPT-4o
In 2024, he coauthors OpenAI’s GPT-4o System Card, which covers the model’s capabilities, limitations, and safety evaluations.
-
2025
Training o3 and o4-mini
In 2025, he leads multimodal post-training for o3 and serves as the directly responsible individual for o4-mini from pretraining through post-training. He leads OpenAI’s Perception team.
-
June 2025
Moving to Meta Superintelligence
Meta hires him from OpenAI for its Superintelligence group in June 2025.
-
2026
Muse Spark multimodal work
At Meta, he serves as multimodal pod lead for Muse Spark in 2026.
-
As of October 2026
Starting a new company
As of October 2026, he is starting a new company after co-founding TBD Lab within Meta Superintelligence Labs.
Sources and credits
Photo: We have no published photo of Jiahui Yu yet. Send us a free-licensed photo.
- UnitBox: An Advanced Object Detection Network
- Free-Form Image Inpainting with Gated Convolution
- Vector-quantized Image Modeling with Improved VQGAN
- Scaling Autoregressive Models for Content-Rich Text-to-Image Generation
- Jiahui Yu
- Gemini: A Family of Highly Capable Multimodal Models
- GPT-4o System Card
- Meta taps four OpenAI researchers for Superintelligence team