Google DeepMind's Thomas Kipf presents multimodal reference conditioning with Gemini Omni at Google I/O, crossing out Transformers on stage in favor of Graph Convolutional Networks

— Technique integrates multiple data modalities for flexible model outputs.

Original post

OPTK#159Thomas Kipf|@TKIPF

Gemini Omni allows me to step into an alternative timeline where Graph Convolutional Nets (GCNs) made it to the big stage 🙃 Jokes aside: excited to finally share how far we've come with multimodal reference conditioning.

11:58 AM · May 19, 2026 View on X

REPLYKD#908Kosta Derpanis (sabbatical in Zurich)|@CSPROFKGD

@tkipf 😂

TKThomas Kipf@tkipf

Gemini Omni allows me to step into an alternative timeline where Graph Convolutional Nets (GCNs) made it to the big stage 🙃 Jokes aside: excited to finally share how far we've come with multimodal reference conditioning.

6:58 PM · May 19, 2026 · 4.3K Views

8:57 PM · May 19, 2026 · 116 Views