InCharacterLive

Inspiration

I've always loved transmigration stories — tales where someone wakes up in a fantasy world and has to figure things out from scratch. One day, I thought: what if the reader could actually be that character? Not just reading, but making choices, hearing the voices, seeing the scenes.

That's how ** InCharacterLive ** was born — an interactive story generator where you provide a premise and the AI brings it to life with text, images, and voiced dialogue.

How I Built It

I started with the core loop: you type an action, and the AI writes what happens next. Then I layered on scene images and multi-speaker audio so each story beat feels like a living moment.

The hardest part was making all these pieces — text, art, voice — work together seamlessly. I added Companion Bot, a real-time voice companion you can talk to mid-story, which turned it from a turn-based tool into something truly conversational.

What I Learned

The biggest lesson was that orchestration is harder than generation. Getting an AI to write a paragraph or draw an image is straightforward. Getting five different AI models to produce a coherent experience together, in a reasonable amount of time, is a completely different challenge.

I spent more time on timing, sequencing, and identity consistency than on any single feature.

Challenges

The generation pipeline originally took over a minute per beat — way too slow. Through profiling and parallelizing steps, I cut that nearly in half.

I also hit subtle bugs: audio that played as silent because it was encoded twice, characters losing their names due to a parsing oversight, and jarring white flashes on scene transitions. Each was a small fix but took patience to track down.

Built With

• fastapi
• gcloud
• google-gemini
• next.js