From Real To Synthetic Humans In Generative AI (JerusML Meetup #44) @Jerusalem

This meetup focuses on the fascinating field of AI personification, where machines breathe life into characters and personas.
**Register here**

Join us as we explore how Generative AI is reshaping the way we perceive and interact with synthetic beings.

📅 When: Tuesday, 09.04.2024 at 18:00
📍 Where: Belgium House, Givaat Ram, Jerusalem
🎙️ Language: All lectures will be held in Hebrew

Agenda:
18:00 - 18:30: Gathering, networking, food and drinks
18:30- 19:10: Can two walk together, unless they are agreed? (Amos 3:3): Compositional image generation with consistent human (and other) characters
Harel Cain - CV Researcher @ Lightricks
19:10 - 19:50: Photorealistic Human Presenters (Agents): Generating Real-Time Personalized Face Animation
Shira Bar On - Deep Learning researcher @ D-ID

More Info:

Can two walk together, unless they are agreed? (Amos 3:3): Compositional image generation with consistent human (and other) characters (Harel Cain)

Abstract:
While the generative AI revolution has taken the world of image and video generation by storm in the last year and a half, some fundamental problems remain without satisfactory simple solutions. Such is the important and practical need to create consistent characters across images and videos, and specifically the problem of placing multiple characters in the same shot with consistent looks and clothing while maintaining natural and sometimes complex interactions between the characters.
We will describe the problem, its causes, some of the latest solutions from academia and share insights into how we tackle this problem in Lightricks. The talk will be based on joint work with Gleb Sterkin.

Bio:
Harel Cain has been at Lightricks for close to seven years. He has led a team of computer vision researchers building the next generation of features for Lightricks products. Since last summer, he has been deeply involved with the new LTX Studio product for generative video storytelling. He was previously a data scientist and security engineer at NDS and Cisco. In his spare time, he translates German literature to Hebrew.
---

Photorealistic Human Presenters (Agents): Generating Real-Time Personalized Face Animation (Shira Bar On)

Abstract:
In recent years, we have been witnessing an exciting revolution in generative models.
A remaining significant challenge in this field is crafting human-like avatars capable of natural interaction.
At D-ID, we are developing cutting-edge generative face animation models that are capable of interacting with users in real time.
Face animation demands techniques that accurately model the human face, allowing for precise control and alteration of its expressions while maintaining the individual's unique identity. Moreover, it necessitates the conversion of audio inputs into corresponding facial expressions, creating a seamless audio-visual alignment. Ultimately, this face model must be synthesized to produce videos that feature convincingly talking avatars, blending realism with responsiveness. This talk will delve into how we fuse all those elements together, with the goal of making them work in real-time.

Bio:
Shira Bar On is a deep learning researcher at D-ID, where she develops the company's cutting-edge talking face models by integrating audio and video generative models. With a B.Sc. in Computer Science from the Hebrew University and a B.A. in Fashion Design from the Bezalel Design Academy, Shira's passion lies in the fusion of design and technology. Her previous experience includes work as a deep learning researcher at Lightricks and internships in fashion design studios in Tel Aviv and New York.