Fake videos of real people -- and how to spot them | Supasorn Suwajanakorn
New Section Which Obama is Real?
In this section, the speaker introduces the concept of creating realistic models of people using existing photos and videos. The goal is to capture their mannerisms and unique characteristics.
Creating Realistic Models
- The speaker was inspired by a project called New Dimensions in Testimony, which allows interactive conversations with a hologram of a real Holocaust survivor.
- The speaker wonders if it's possible to create similar models for anyone, including historical figures or loved ones.
- A new technique is introduced to reconstruct a high-detailed 3D face model from any image without 3D-scanning the person.
- Analyzing a large photo collection of the person beforehand is key to building an accurate model.
- Controllable models can be created from internet photos, allowing for motion transfer and driving the model with videos.
- A neural network is used to convert audio into mouth points, synthesizing realistic speech movements.
Ethical Considerations
- The speaker acknowledges that while the results are realistic and intriguing, there is also a potential for misrepresentation. The goal was to build an accurate model of a person, not to deceive or misrepresent them.
New Section Conclusion
This section highlights the possibilities and challenges of creating realistic models of people using existing media. It also raises ethical considerations regarding representation and consent.
Key Takeaways
- Creating realistic models using existing photos and videos opens up creative possibilities such as preserving historical figures' teachings or allowing loved ones to give advice even after they have passed away.
- The process involves reconstructing 3D face models, analyzing photo collections, and synthesizing speech movements.
- Ethical considerations include the potential for misrepresentation and the importance of consent when creating models of individuals.
The transcript provided does not contain a conclusion section. Therefore, the summary is based on the available content.
New Section
People have been thinking about the problem of fake images and videos since the early days of Photoshop. The speaker is a researcher working on countermeasure technology at AI Foundation. They mention a tool called Reality Defender that can automatically flag potentially fake content in web browsers. However, they emphasize the importance of being critical and aware of what we see, as there is still a long way to go in fully modeling individuals and ensuring the safety of this technology.
Fake Content Detection Efforts
- As a researcher at AI Foundation, the speaker is involved in using machine learning and human moderators to detect fake images and videos.
- They mention an ongoing effort to develop countermeasure technology.
- One tool they plan to release is called Reality Defender, which is a web-browser plug-in that can automatically flag potentially fake content.
- Despite these efforts, fake videos can still cause damage before verification takes place.
Importance of Awareness and Critical Thinking
- It is crucial for everyone to be aware of what is currently possible with fake videos.
- Being critical about what we see helps us make accurate assumptions.
- Fully modeling individual people and ensuring the safety of this technology are still significant challenges that need to be addressed.
New Section
The speaker acknowledges that there is still much progress needed before fully modeling individuals becomes possible and before ensuring the safety of this technology.
Challenges Ahead
- There is still a long way to go in fully modeling individual people using this technology.
- Ensuring the safety of this technology remains an important concern.