[LA27] - 360Fashion Maze Game

Inspiration

This game is designed to overcome bias through gameplay and fashion by presenting models of all race and size, thereby championing true beauty and changing our relationship to it.

As a former fashion model Anina understands what it is to be in a fashion show. As one of the worlds first mobile fashion bloggers she understands what it is to run from show to show to get the shots for your social media. As a fashion technologist she understands the latest technology and how it can be applied to fashion. As a gamer she understands how games are primarily made by men and there are few games for girls. As a woman she understands how debilitating it is to try to fit into a body size the world says is beautiful. As hackers we understand how we can change perceptions and realities through narratives, visuals, and technology.

What it does

World Model action-video generation of Fashion Show scenes, one after another. Gameplay: Race against the clock to make it in time to sit in the front row of the Grand Finale Fashion Show. All along the way you’ll find couture fashion installations that wow and amaze you. Snap a photo, and move on or you’ll arrive just as the gates are closing!

How we built it

codex with gpt 5.6 sol ultra mode, under parallel sessions coordination harness. Triplo GPT-Image-2 for character design and extraction from midjourney photos.

Challenges we ran into

Consistency issue. Out of control visuals drifting. Reminds me of the very early days of LLM and image/ 3d generation. Lots of hallucinations and disobey. As the ncp’s data is trained and are designed for shooter games, they behaved in a shooter way. Moving, crouching, even some models lifting their arms to shoot, nervous, and turning their backs to the camera no matter how many times we tried to prompt it out of them. Additionally because of the data set they are trained on, our diverse characters degraded into thin white characters that had nothing to do with our rich and diverse size and culture initial models. The engine also invented bystanders in normal clothing outfits that had no inspiration from our amazing fashion designs. It was almost impossible to get the character to turn towards the camera, they constantly huddled together and came very close to the camera looking nervous and uncomfortable and Caucasian.

Accomplishments that we're proud of

Scene transitions. GPT Imagine 2 generated final scene (no one knows what's next in fashion). Camera operation. Reactor SDK deployment. The beauty of the initial scenes from the first frames.

What we learned

Video-action centric World Model still has a long way to go. Data sets need wider training data and diverse teams or we can not create an inclusive world.

What's next for [LA27] - 360Fashion Maze Game

Looking forward to swap out the Lingbot2 with more advanced ABot-World and DreamX-World models on reactor. Hopefully an effortless migration.

Built With

• claude
• codex
• html5
• javascript
• reactor
• triplo