Deepfake Technology Live Demonstration

Overview

The speaker discusses new software that enables streaming deep fakes in real time, specifically mentioning Deep Face Live.
Comparison to previous depict videos which required significant processing time.

Deep Face Live allows live use of deep fake models.
Ability to impersonate personalities like Jim Carrey and Sylvester Stallone in real time.
The fun aspect of using the technology is noted, comparing it to playing VR.

The speaker plans to prank co-workers by using Deep Face Live to assume the identity of a colleague named Ren.
Intends to create a perfect deep fake model of Ren to deceive others.

Data Collection:
- Collect various examples of Ren’s face under different conditions (expressions, angles, lighting).
- Use sentences that contain every phoneme to capture a variety of mouth shapes.
Training the AI:
- Initial training on random faces to teach the AI facial structure basics.
- Gradually replace random faces with Ren’s face.
- Final stage includes training the model on the speaker's own face to refine expressions.
Machine Setup:
- Utilizes a machine with three 30 90 GPUs for training which will take approximately two weeks.

ExpressVPN:
- Mentioned as a sponsorship. Provides internet encryption and allows access to geo-restricted content.

The speaker demonstrates the deep fake on co-workers, receiving mixed reactions.
- Some colleagues quickly identified the deception, while others were initially confused.
- Notable feedback includes observations about the difference in facial expressions and overall impressions.

Discussion on the effectiveness of the deep fake technology and the possibility it holds.
Reflection on voice matching as a future challenge for deeper immersion.
Plans for further experiments, possibly impersonating a celebrity on platforms like Omegle.

Excitement for the capabilities of deep fake technology, mixed with caution about its implications.
Encouragement to subscribe for future experiments with deep fakes.