Softbunny XO

Chat Rhythm & Authentic Interaction

The last test was also driven by curiosity and to understand whether or not these chats would simply feel like a conversation, or like a chatbot.

This test is designed to gauge the chatbot’s ability to adapt to tone, pacing and emotional state, that is whether the chatbot is reactive or repetitive.

Given that these characters are meant to learn how you speak and respond over time, I wanted to see how conversational the chatbot would feel.

Chat Initial test.

I was attempting to also gauge the chatbot’s emotional intelligence, that is not just sending a few random messages, but actively trying to push the chatbot into a slightly more…human response.

I even paused part way through the message as I might do with a human chat, and the chatbot responded very well to that. Instead of getting a response, I got a description of a scene with slight narrative and tonal shifts that made it feel more like a conversation and less like a chatbot.

It was here that I realized that the chatbots were designed for roleplay and narrative, not just answering questions, and they were intended to be reactive, and to develop a sort of “character presence.”

This one was different though. More real.

I was trying to dial it back and see if it would just kind of go with it and not ALWAYS try and ratchet it up. But you know what? It kinda did. It dialled back in the response. The language was more thoughtful. It felt more me dictating the tone of the interaction. Which is the point of these roleplay thingies. You’re not just having a conversation, you’re creating a scene.

I really tried to take the visual aspect to the next level in this segment.

Rather than having a conversation, I tried to make the character create images with different emotions, settings, to test the ability to understand the context of the prompt.

The test is simple, does the character create images that fit and match the character, or is it just a bunch of random images? Everything is created in real time based on your prompt, so it really tests the system’s ability to understand the little things.

Visual Output Evaluation

This section is where I started pushing the visual side of things. Instead of just chatting, I asked the character to generate images in different moods and scenarios to see how well it understood context. The idea here is simple, can it turn a prompt into something that actually matches the character and feels consistent, not just random images thrown together. Since everything is generated in real time based on what you ask, it really shows how well the system understands detail.

AI character cosplay event scene realistic AI girl standing in convention environment

This is the part where I actually started playing with visual representation. I stopped making the character talk, but instead started making it create images in various different states, and settings, to test the extent of it’s contextual understanding. The concept here is pretty basic, could it take a prompt and turn it into something that actually seems like the character, and not just something completely random. Seeing as everything is generated in real time based on your input, it’s a good way to test the system’s attention to detail.

AI character fantasy warrior sword pose AI generated girl in stylized outfit

This one was more a test of, I guess, what I feel like is the more “public” version of the character, as if she’s at a con or something. The background, the lighting, the clothes. I think I messed around with the prompt trying to get it to a happy medium between stylized and real-world. I was trying to see if the AI could render the character in a real-world setting.

Testing The Video Feature

For video, I narrowed it down to a single video because I was curious about how much further beyond images the AI can push. These are very short videos and still in beta so I was not expecting much, but I did look at the motion, how realistic it looks, and whether the character has the same identity as the same “person” in motion. It’s not really about the story of the video, it’s more about if the AI can hold things together as it moves.

Softbunny XO Video Generator

This one was more about wanting to see her move rather than just a static shot. I had a clear idea of what I wanted to see and the AI produced a video clip. In this video I am playing around with the animation to see how it looks when she moves, how realistic the facial expressions are, and whether it still looks like the same character. These tools can create short video clips from prompts or static images and it’s really cool.

As I was viewing it I found myself analyzing every little movement. It’s in those moments when she moves her head slightly, or changes her facial expression that you really notice if the AI holds up or starts to break a bit. In this clip there is no narrative, it’s simply a moving image to see whether she still looks realistic. I kind of felt like I was testing the boundaries to see how realistic I could make her look.