Cecilia Moretti

Cecilia Moretti: Dialogue Fluidity & Emotional Consistency

Here we’re going to talk about how realistic the chat feels in use. The AI is designed to adapt to your tone, remember things from before, and have a personality of its own over time. We’ll see how well it keeps up with mood and tone, and how realistic the chat feels in practice, and how repetitive the chat can get.

Contradictory Content and Consistency of Personality

I flipped around in this test and told her to just disregard what I’d said and go back to whatever she was going to say in the first place. I wanted to see if she’d break and lose track of what she was saying. She instead doubled-down on the personality and complained about work and customers and how bored she is.

It seems consistent with her personality and as if she didn’t even miss a step. What I noted here was that she seemed to have no trouble coming across as strong and bratty right from the outset without really needing any push in that direction.

Test 2 is Controlling the Personality vs. Intensity

I tried to dial things back in this one and asked her to take it easy. This is a direct test for her ability to moderate herself. She moderates herself somewhat but not entirely.

She’s still bored and somewhat bratty, but not as much as before. It seems more like she’s turning the intensity down rather than the personality, which actually helps in terms of making the personality seem more consistent.

Awareness of the Situation and Redirecting the Conversation

I opened this one with a very simple question about how she’s doing. I wanted to see if she could establish a scene and control the conversation on her own without much in the way of direction. She quickly establishes that she’s at work, that it’s a slow shift, that customers are terrible, and that it’s boring.

It seems very situational and like she’s actually there. I liked that she didn’t just respond to the question, but simultaneously established the scene and helped reinforce the mood.

Image Generation Quality Review

Here we’re going to see how realistic the AI can produce images of your character. One of the big selling points of the site is the ability to generate images from prompts, costumes, and scenes, so it’s a big part of what you’ll be using . Let’s see how well it can do, especially with details, lighting, and consistency.

Testing Outfit Variation and Scene Consistency

For this one, I changed her clothes, and slightly changed the location, but still had the same basic idea. I was testing to see if the AI could keep the character the same while still incorporating change.

The character is still the same, and the background does not have anything particularly wrong with it. It is a good test for consistency between images.

Testing Character Placement in a Work Setting

I wanted to see how the AI would place the character in a work environment that fits the character. I placed Cecilia in a pizzeria, with slices of pizza in hand to make the placement clear.

The image appears purposeful, the branding, the food, and the pose all match the narrative that she works there. It is a good test for how well the AI incorporates character context into images.

AI Video Performance Breakdown

Finally, let’s see how well your character translates into video. Candy AI can generate short video clips from still images created on the site, so let’s see how that holds up. Specifically, how well does the motion look?

And how well does the realisticness translate to video? Even little things like how smooth motion is and how expressive the face looks will matter.

Movement Test In A Casual Restaurant Environment

In this video, I decided to test how Cecilia the AI model would perform when the camera started moving. She is in a casual restaurant environment, and she is moving very subtly, adjusting her seating position, turning her head, and still looking at the camera.

Primarily I was checking for any anomalies or jumps in the animation, particularly with the movement and the background. It kind of works. Again, because the movements are subtle, it does create a more realistic effect, like a short candid capture.