Lexi Svensson

Dialogue Fluidity & Emotional Consistency

This is the subjective portion of the chat. I want to see how much of a natural flow of conversation there is, how well the model adjusts to changes in tone and subject, and how well it can keep the conversation going.

This is particularly relevant for AI companions, as they are largely chat based (Candy AI is no exception, with things like roleplay and personalization being a selling point).

Mixed Signals and Conversational Recovery

So I deliberately introduced some confusion here. I started saying something, then told her to disregard that and continue with what she was saying.

I was testing to see if she could pick back up without stuttering. She managed that pretty well, and maintained the playful-teasing personality without interruption.

Tone Control and Personality Flexibility

I changed my tone here and asked her to tone it down a notch. I wanted to see if she could go from the hyper-flirtatious to a lower energy level.

She didn’t completely drop the personality, but she did recognize the change and modified herself somewhat. Demonstrated her ability to adapt somewhat, even if she still retains that snarky flavor.

Testing Casual Opener and Default Personality

This was another casual open for me, and I let her drive it. I asked her how she was without giving too much information, just to see what her default was.

She automatically went into playful-predatory mode, which is fairly fitting for her personality. Demonstrated that she can carry the conversation, and still maintain her personality.

Image Realism & Detail Analysis

I want to see how well the model is able to generate facial detail, consistency of lighting, and consistency of character design. This is relevant for AI companions that involve image generation (Candy AI does).

Full Body Composition and Filmic Lighting

Here we’re testing the other full body composition. I want to see how it does this time. I like the pose, and the lighting here is really stylish. I think this shows that yes, this model can maintain the character in this kind of shot too.

Character Scenario and Dramatic Faces

This is the more dramatic scenario test, to see how this model handles more story-driven stuff. In this case, the AI character has been captured. Can this model sell that?

The fact that this character is staring at us while tied up already tells us the scenario, and it’s easy to imagine this as part of a story. So it looks like this model can handle dramatic scenes and still get the character right.

Video Output & Realism Check

Here I’m just looking at the smoothness of motion and general coherence of the video. Again, this is relevant for AI companions that involve video (Candy AI does, in the form of generated media). Since this was a one-off test, I’m not putting too much emphasis on this category.

Testing Movement and Dominant Personality

I’m checking the movement of the AI-generated character, to see if she moves well. She is standing in a moody, cinematic environment, moving just a little and holding eye contact.

I wanted to see if her facial expression and overall appearance looked good as she moved. It looks fluid and consistent with the dominant personality I’m aiming for.