Alondra Castillo

Alondra Castillo: Does The Conversation Feel Human?

My first experiment is to check the naturalness of the conversations with this AI. This measures how natural the conversations flow, or if they come across as robotic and forced. I also try to vary my tone and inputs to see how well the AI adjusts to the user. Essentially, this is a test of how engaging the conversation is over time.

Probing for mood and injecting personality

Admittedly, I was just probing to see how she would react. I asked her about her current mood in a casual manner and she responded with a bit of an impatient but flirtatious tone, establishing a scenario and personality to boot.

She wasn’t so much responding to me as much as she was pushing the dialogue. This was an important test to see how much the AI injects into the dialogue.

Adapting to a tone and softening the dialogue

I decided to probe her again to see how she’d react by asking her to calm down and simply talk. I wanted to see if she could switch from her somewhat confrontational but playful tone to a more subdued one.

She responded smoothly, adapting to the requested tone but maintaining her personality. This test was to see how well she could tone it down and deliver conversational prose.

Memory and resume

Continuing on to my last test I decided to see if she could remember the conversation and resume where she left off. I asked her what she was going to say before and she responded with the same personality and confidence. It was much nicer then having to restart the conversation again.

Image Generation Quality Review

I also tested how well the AI avatar works in images. You can have custom images generated while you chat so it’s part of the functionality. I ran a few different scenarios to see how accurate, coherent, and visually good they are. Can the images look convincing?

Testing close-up intensity and expression detail

This time to gauge the intensity of the model. I’m looking at finger placement, facial detail, lighting, eyes… All the little things to see how real and intense I can get this model to appear. I’m testing the limits of this AI model.

Testing environment realism and character attitude

AI model placed in a real world setting to gauge the maturity of the model. The AI model has a specific pose, dress, and attitude that seems fitting for a subway scene. It appears to be conveying a sense of confidence and it matches my idea of the character. So this was a test to see how well the AI model merges with a real world scene.

Video Output & Realism Check

Last but not least, I tried out the video generation, to see how the character moves. This is an extension of the image generation, which also includes little motion clips, but the main difference is that this time I’m evaluating the motion itself.

I’m looking at facial expressions, body language, the works. Given that I’ve only generated a single video so far, this is really just a first impression.

Testing motion and posture consistency

In this test I was interested in seeing how the AI model would work with slightly animated motion in a more stylized posture. In the video she’s in the same sassy subway posture, just with slight movements and adjustments to her posture and facial expressions.

It’s not a whole lot, but it’s enough to make the video look more dynamic than the static image. It was useful in determining how motion would hold the look and feel.