69th Vienna Deep Learning Meetup: Real-time Video with Diffusion Models

Hi Deep Learners,

We are announcing our next Deep Learning Meetup and are glad to have Rahim Entezari from Stability AI presenting Real-time Video Generation with Diffusion Models and Florian Kowarsch from nuseum discussing How Mixture of Experts (MoE) Actually Works.

***
Agenda:
18:45 (note the 15 min later start than usual)

• Introduction & Welcome by the meetup organizers & hosts

19:00

• Talk 1: Real-time Video Generation with Diffusion Models
  by Rahim Entezari (Stability AI)

19:45

• Talk 2: How Mixture of Experts (MoE) Actually Works
  by Florian Kowarsch (nuseum)

20:15

• Announcements
• Networking & Discussion

~ 22:00 Wrap up & End
***

Talk Details:
Talk 1:
Real-time video generation is rapidly emerging as a transformative frontier in generative AI, pushing beyond static imagery to dynamic, interactive media. This talk explores recent advances in diffusion and transformer-based architectures that enable fast, high-quality video synthesis, highlighting innovations in sparse attention, distillation, and autoregressive generation. Rahim will discuss the technical challenges of achieving both speed and fidelity at scale, and showcase how these breakthroughs open the door to new creative, interactive, and production-ready applications.

About the Speaker:
Dr. Rahim Entezari is a Research Scientist at Stability AI, where he actively contributes in advancing large-scale generative models for image and video. He was a core contributor to Stable Diffusion 3 and 3.5, and is currently driving research on real-time video generation. Before that he earned his PhD in Artificial Intelligence from TU Graz with distinction.

Talk 2:
A central trade-off in training these models is balancing two competing priorities: how specialized each expert becomes and how evenly the network distributes computation among them. In this talk, Florian discusses how the implementation of MoE in major LLMs might seem counterintuitive at first—but makes perfect sense when viewed through the lens of this trade-off.

About the Speaker:
Florian Kowarsch studied Data Science at TU Wien and worked for 2 years at a research project at the TU Wien Computer Vision Lab. After moving to Singapore to work for a Face Recognition startup, he co-founded a small company that produces AI-based audioguide solutions for museums.

We are very much looking forward to seeing you at our next meetup!
Your VDLM organizers