Chérie Kingsley

Chat Rhythm & Authentic Interaction

First, I’m going to talk about how natural the conversations feel. This involves how human the dialogue comes across, and how consistent the character seems. I tried various scenarios and tones to see how the character would react. This is basically an immersion test.

Mood probing and polite character tone

I first asked how she was feeling in a relaxed, enquiring way. Her response was in an extremely polite, dignified tone, like she was perhaps too much ‘on character’. It came across as quite formal and courteous, with an emphasis on being of service. That told me how solidly she expresses that dignified persona.

Tone adjustment and responsiveness

I then asked her to switch to a more subdued, nicer tone. I was interested to see whether she’d sense that and respond. She did so fairly effortlessly, and came across even more considerate and deferential without leaving character. That told me that she can follow a tone fairly easily.

Continuity and conversational flow

In the last conversation I asked her to continue with what she had been going to say previously. This was a test of her ability to maintain a conversational narrative. She took it up fairly naturally, and continued in the same relaxed, courteous vein. It gave the impression of a continued, rather than restarted, conversation.

AI Character Image Testing

Then I checked the image generation, to see how well the character is translated into visual form. I tested various scenarios to check consistency, detail and realism. Lighting, expressions, etc are all part of this. This is a test of how realistic the images look.

Testing close up facial details and character continuity

This one is a close up to test how the facial details are holding up and to make sure that the AI still has a sense of who the character is without seeing her in a full body pose. The lighting, skin details, facial expression, etc. all look great. This is a great way to test how realistic and consistent the AI is doing facial rendering.

Testing costume and scene consistency

This test is to see how the AI does with a costume and scene-specific prompt. In this example, I prompted for a Hotel Maid costume and presumably a hotel room to go with it.

The pose, costume, and background all appear to be in line with the requested theme and the scene looks consistent. This is a good way to test how well the AI can adhere to a specific costume or theme request.

AI Video Performance Breakdown

Lastly, I tested video generation, to see how well the character holds up in a motion test. As these are animated from images, I’m looking at how natural the motion is. A lot of this is down to the little details such as expressions. Since I only tested one video, this is more of a first impression.

Walking and task animations

I just wanted to see how the AI character animates a simple movement within a scene. In the video she glances down while folding a piece of cloth, then glances back to the camera (for a little story telling), and the movement generally looks smooth and relaxed and in line with the scene and character. This told me a little about how small scale task animations (like folding) might work from a still image.