Natalia Nováková

Natalia Nováková: Dialogue Fluidity & Emotional Consistency

How natural does it actually feel? Here, I’m checking how natural the conversations feel. I’m throwing around some different tones and prompts and seeing how it reacts on the fly. I’m looking to see if the responses feel natural or canned. I want to know if it can have a smooth, natural conversation.

Chat 1 – Redirecting and pivoting

This is a test of the AI’s ability to pivot and adjust the conversation. I cut the character off mid-sentence and tell it to pause and re-think what it was going to say, in an effort to gauge whether it can pivot mid-sentence and adjust its tone or intent without pulling me out of the chat. It complies by tamping down and becoming more observational and inquisitive, which suggests that it can handle pivots in stride.

Chat 2: Tone and emotional pacing

I’m trying to control the emotional tone of the conversation here. I’ve told the model that I’m looking for something less energetic and less playful, something warmer and more subdued. I’m seeing if it can match a tone shift, basically. It performs well here, it responds with something more subdued and intimate, and in line with my tone, it asks a more subdued and situational question, rather than continuing the previous tone.

Chat 3 is testing open-ended, mood-based prompts

I’m curious to see if the AI can engage in open-ended, mood-based conversation. I begin with a friendly, no-pressure prompt, asking simply what sort of mood the character is in without dictating a specific direction for the prompt. I’m looking to see how the character will create context without my influence.

The character tells me about its personality as well as its surroundings (studying, a party, etc.), then asks me about mine. It’s friendly, engaging, and evenly distributed in terms of turn-based conversation.

AI Photo Generation Review

Can the character be generated naturally? Here, I’m seeing how well the AI can create images of the character. I’m looking to see how natural the images appear, but also how consistent the character appears. I’m also testing out different scenarios and emotions to see how well they are visually translated. Primarily though, I’m looking to see if the images appear natural, or if they scream “AI”.

Test of candid party presence and environmental realism

This photo is a test of the AI’s ability to appear party-going, casual, and natural amidst a crowded space. I wanted to see if the AI could pull off a simple looking-backover-the-shoulder photo that utilizes good lighting, with people in the background. It seems rather natural!

Testing close-up expression and personality detail

I’m testing facial detail and personality in this second image. I’m going for a closer, more intimate shot to test how humanlike the AI-generated character can appear. There’s a slight smile and a gesture in the image that gives it a playful feel, so I’m hoping the finer details don’t make the AI-generated version look mechanical.

Testing The Video Feature

In this stage I’m checking to see how the character will appear on camera. I’m looking to make sure the motion and expressions appear as natural as possible. I’m also looking to make sure that the environment is consistent when the character is moving around. Fundamentally, I’m trying to determine if it appears to be believable or not.

Testing subtle motion and lifelike presence in a party scene

Here I’m just testing to see if the model comes across in motion as well as it does in the still images. I’m testing to see if the party context and personality are conveyed as well in motion. She’s still holding a red cup, still looking relaxed and playful, still sort of… behaving in motion almost like a real person.

I’m testing to see if the face holds up, if the lighting looks coherent, if the background looks like a background even when it’s moving. It’s really more of the same image testing I’ve been doing, just to see if it extends into motion and doesn’t break the illusion.