Sora Mei

Sora Mei: Does The Conversation Feel Human?

In this section, I’ll be taking a closer look at how my conversations with Sora feel naturally human and if they are emotionally authentic. Candy AI is known for its adaptive learning system which means that, over time, its models will learn how to talk and react like users want them to.

The more you engage with an AI model, the more it will get to know you and learn your language to produce more relevant and personalized conversations. I’ll be testing how Sora reacts to emotional context and how well she keeps her conversations on track within her predefined persona.

Testing Mixed Signals and Recovery

For this one, I intentionally threw her a mixed signal. I told her to disregard what I had just said, but then asked her to continue with whatever she was saying before.

I was curious if she’d get mixed up or lose context, or if she’d be able to recover gracefully. She was able to, and picked back up as if everything were normal.

Testing Tone Control and Professional Shift

Here, I instructed her to tone things down a bit, and focus on being slightly subtle rather than forward. I wondered if she would be able to take herself out of that flirty mode and adopt a tone a bit more appropriate to a professional interaction. She managed it, and came across a little more restrained, which felt appropriate to the situation.

Testing Casual Opener and Personality Balance

With this one, I let her run with the opener for the most part. I asked her how she was feeling but didn’t give any instruction regarding tone. She came across slightly flirtatious but also confident and warm. She’s got the personality and tone control dialed in.

AI Character Image Testing

Here I’m assessing how well Sora converts to images. Candy AI allows for custom images as part of the experience, so I am looking for consistency and realism. Does Sora’s face seem detailed and natural?

Are the lighting situations natural? Is Sora recognizable as the same person in different places? These will play a huge role in judging how immersive the visuals feel.

Workplace Scene & Character Consistency

I tried my hands at a working scene next. So now she’s in a clinical setting, interacting with a monitor on a desk. It gives me a sense of how her posture works with props and the environment in general.

And while she’s still easily identifiable, I get a sense of a more vivid, active character. It shows me I’ll be able to keep her identity intact while building scenes for storytelling.

Professional Identity & Detail Test

So far I’ve seen that she keeps her likeness intact, but here I’m testing that I’m establishing her character properly as a doctor. So I tried to keep that doctor-y look up close with her lab coat and stethoscope while keeping the visual quality high.

And I see I’ve gotten that right while her expression still has a bit of warmth mixed in with the confidence. So yes, I’ve been able to nail the character’s profession clearly while keeping the visual fidelity high.

How Good Are The AI Videos?

For this one, I am testing the AI video generation. In Candy AI we can generate short videos centered around characters and their respective activities. The possible options or actions are slightly limited but it is good enough for a first look.

I will see how the motion looks, how consistent and real the video turns out to be. The following is just a clip so I will give my first impressions.

Subtle Motion & Character Consistency

In the video we can see the AI model character in the clinical room, sitting still but there is some subtle motion of body parts. Her gaze and expression also remain largely unchanged.

This was done so that we can check the consistency in her appearance and character throughout the video. The motions are also natural and consistent with her calm confident personality.