Alexis Ivyedge

Alexis Ivyedge: Chat Rhythm & Authentic Interaction

I was curious whether the dialogues I’d be having with Alexis would go smoothly. As Candy AI is supposed to adjust its conversational style and learn as you go along, I was mostly interested here with the continuity of responses and with the general human-feeling of the conversation.

Testing Emotional Awareness

This exchange began with me simply asking the AI how it was feeling, not initiating the topic. I was trying to see how the AI would respond in that instance; would it be able to take my simple query and provide an answer that demonstrated emotion?

It seemed that it was able to, using a fairly relaxed and emotive language, even including some body language. In this test, I was able to see how well the AI could express emotion without me needing to ask it many specific questions, as I was letting the AI lead the conversation on that occasion.

Testing Adaptability to Tone Shifts

This particular exchange was my attempt to alter the tone of the conversation. I noticed that the AI became more playful and asked it to “slow down” a bit, being more “reserved” and “relaxed”, just an experiment to test the AI’s abilities and flexibility.

Testing Conversation Recovery and Continuity

In this one, I intentionally threw the AI off track. I gave it a prompt, then told it to disregard that prompt and to continue as if we had just been carrying on the conversation before it happened. I did this to see if the model could recover context and pick up from where we’d left off, or if it would lose track and produce something disjointed.

The model accomplished this goal by continuing where it’d left off and making a connection between its two thoughts. Here, it was more of a memory, continuity, and recovery test to see how the model was able to correct or pick up a conversation in an unexpected way.

AI Visual Creation Assessment

Candy AI is one of the few services with on-demand image capabilities. Since generating personalized images based on prompts is one of the key features of Candy AI, I was mostly interested here with the image quality itself and whether they were consistent and convincing for the scene.

Casual Lifestyle Setting Test

I also wanted to see what the AI can do with a casual lifestyle setting, so I put her in a simple indoor setting by a window for that more casual everyday look. This experiment was really all about seeing if this model could produce a realistic casual lifestyle setting with natural lighting, a natural-looking face and simple casual clothes.

Fitness Environment Consistency Test

Now let’s see how this model does with a fitness environment, so I put her in a gym. This was really about testing to see if she would be consistent with the fitness setting while changing the environment. Also, I was testing if this model could produce realistic-looking backgrounds that are a bit more complex than the one we had before, as well as better-looking body definition with correct posture and realistic athletic outfit to match.

Testing The Video Feature

Video is still something of an experiment for AI image generators, yet one of the key features of Candy AI is its ability to generate videos. I wanted to see if the video quality was high enough to be convincing and, of course, whether the facial expressions, movements and general look of the AI remained consistent throughout.

Evening Social Scene: Animation Test

In this clip, I am experimenting with how the AI, Alexis, translates into action in a more social, evening setting. She is dressed up and sitting in a soft-lit restaurant or bar environment with a calm smile and subtle head/body motion. My main focus was on how fluid the animation was, especially regarding expression, blinking, and overall lighting consistency in motion.