Beatriz Vargaro

Does The Conversation Feel Human?

Chat is the main feature I looked at because, at the end of the day, a companion AI is a conversational AI designed to replicate a real relationship, so I wanted to test the conversational AI. I test how well it responds to changes in moods and tones, how consistent it is, etc. The aim is to get a feel for how natural and human-like the chat function is.

Testing mood detection and dominant personality

I began with a fairly broad question about how she was feeling to gauge her response. She gave a very confident and slightly dominant answer, throwing in some extra attitude and embellishments.

It was quite characterful and a bit more like acting out a character than actually answering the question. This was helpful to test the extent to which she can maintain a bold personality when left to her own devices.

Testing tone down-shift & flexibility

I then purposefully changed my tone and asked her to tone it down a notch. I was curious whether she could shift from the sassy banter to a softer more chill tone. She more or less succeeded (even if she had to maintain some level of her personality). That’s a plus, since it means the AI can adjust without totally falling apart.

Testing continuity and playful control

We ended the previous conversation and I allowed her to continue with what she was going to say. This test was more about continuity and seeing if she could remain playful.

She continued with the same playful tone and even connected it to a little “mission” which added some variety. It allowed me to see if she could sustain things from one message to the next.

Image Realism & Detail Analysis

The second main thing I tested was the image generation, because the service markets itself on customisability and visual fidelity, so I tested that. I wanted to see how different scenarios and environments would affect the image quality, how the lighting and the consistency looked, etc. I also compare how realistic the model looked compared to real life.

Testing pose variety and personality expression

With this next one, I wanted to push the personality and posing a little further and see what the limits were on this image generation. I produced a much more playful, almost campy pose in a similar studio setting. The output is much more stylized and less formal.

This is just another example of the more expressive results the AI can produce, and another example of how the character can be much more professional or much more personality based on the prompt.

Testing studio realism and presentation style

I wanted to test how the AI model would perform in a more professional setting, so I chose a weather studio backdrop for this prompt. Here, Beatriz AI character appears in the midst of a weather report, with the graphics behind her reinforcing the setting.

The lighting and shine on this image give it a live-on-air feel. I was curious to see how the AI model would merge the AI character with the background.

How Good Are The AI Videos?

The last thing I tested was the video generation, and I am aware that most services generate videos simply by adding motion to an image, so I tested that. I tested the motion, the expressions, how realistic it looked compared to the images, etc. Since I’ve only tested one video, I will admit that this may not be representative.

Motion and subtle character animation

I used this test to assess the quality of the AI character when a static frame is combined with a little motion. In the video, she swings her head around, lowers and raises her eyes, and aims them at the camera, making the animation more believable.

She’s got a little gesture when she moves, some personality, but she hasn’t fundamentally altered the composition of the original frame. It gave me an idea of the quality of small motions applied to a generated image.