Elodie Valmont

Dialogue Fluidity & Emotional Consistency

First, I wanted to see how intuitive the interactions are. I used a variety of inputs, styles, and emotions to gauge the ability of the chatbot to match the tone and maintain the flow.

The questions I needed to answer were whether the output is natural or if it becomes monotonous and robotic at any point, and also, how consistent is the personality throughout the conversation.

Questioning emotional depth and reflection

I wanted to challenge the AI to something slightly more profound. I did not attempt to lead the AI in any way but simply asked it what it was thinking about.

The response was quite interesting as it involved turning 18, becoming an adult, and preparing to go to university. I believe this was a fair test of the AI’s ability to share a mix of emotions (excitement and nerves) in an unstructured way.

Trying personality and situational awareness

In the last test, I attempted to measure personality and tone. I asked her “how do you feel?” I was hoping for a more energetic description of her current state and she did not disappoint. “I’m excited, nervous, and slightly drunk,” she replied.

She provided some context for her answer by saying she had already had a few drinks before the party with some friends. More energetic, more situational, in the midst of something. I used this test to see how the AI would simulate a live environment and social situation.

Visual Output Evaluation

Then I tested the image generation, to check how the AI character will be presented in various situations. I generated a few different scenarios to evaluate the visual impression and consistency. Lighting, facial features, environment, and other factors are all part of this.

And of course, I verified if the images fit the identity and personality that was established through text chat.

Testing social interactions

I was curious to see how the AI model would behave in a social setting. In this example, I decided to generate a selfie of the AI model during a birthday party to see whether it could capture her personality. The background, the decorations, the posture, etc, all help with the believability of the scene.

Testing the birthday scene

I decided to test the ability of the AI model to generate more close-up, emotional scenes. In the generated image, the AI model is blowing out the candles from her birthday cake (18th birthday). This image looks fairly realistic (the lighting, the facial expression, the candles, etc), and the scene looks intimate.

Video Output & Realism Check

Lastly, I tried out the video generation to see the performance of the AI character. Here, I just wanted to check if the movements and facial expressions are smooth and natural.

Essentially, I wanted to see if the character still looks “real” when it is no longer a static photo. Given that I only tried one video, this review is more of a first impression.

Motion and character realism in video

I now try and see how the character holds up in motion. I create a small video of the character behaving in front of the camera as if I just took a selfie together for her birthday. The character feels more real with these minute expressions and motions, compared to just a static image, which is what I’m testing.