Suryakant Sahoo

Motivation (Why)

As part of our course on Experiments with AI-Generated Media, we were posed with a pro-level challenge - to use the first-order models for video generation and create a deepfake of our younger/older self talking back to our present self!
I saw this as a good opportunity to expose myself to the nuances of GAN while clubbing multiple models to achieve the goal.

---

Techtics (What & How)

Creating the young self seems pretty simple and straightforward, would need to use a photograph of my younger version. But I wanted to have an adventure and make it a bit challenging, so I ventured into digitally aging myself and then making that version talkback from the future!

For Aging myself, used the Lifespan Age Transformation Synthesis trained GAN model to de-age and age myself from my current photograph.
Then used the last frame of this DeepFake Old myself and applied it to the First Order GAN model as an input to manipulate a static digitally aged photograph of mine using a recorded video with facial movements.
Finally, crafted the resulting video overlaying the generated deepfake video with an audio 'Message from the Future' of old me talking to the young/current me.

---

Reflection & Application (now & next)

This micro-project was undertaken to learn and have fun, the results were awesome, while a bit freaky as to think that this level of manipulation can be done in current times and this is going to become more flawless in the future.
I imagined many future applications of this amazing technology, Animators can voice the characters without the need to draw additional frames for animations. Educators can animate Scientists, Emperors, and Politicians to encourage students to study Science, Math, History, Politics, and maybe make them aware and request to actively work to reduce Global Warming. Psychotherapist can use this technique to guide children and adults to promote self-confidence, etc.

---

Result Videos

---

Tools & Technology

---

Creators

Suryakant Sahoo