Dreamwear | Devpost

Dreamwear's home page
The wardrobe
Generating a new design

Inspiration

The inspiration for Dreamwear came from a desire to bridge the gap between creativity and technology. We noticed a growing trend in personalization and self-expression through clothing. People wanted to wear designs that represented their unique personalities and interests, but they often struggled with the design process.

What it does

Dreamwear combines AI-generated designs with AR technology to streamline the design process and provide an engaging, interactive experience. You can create unique designs for a variety of articles of clothing using a single text prompt, save the design to your virtual closet, and try on infinite new outfits; all in real time.

How we built it

We began with the front-end (using SvelteKit, Threlte - a wrapper for Three.js in Svelte, and Tailwind), testing Stable Diffusion for design generation, and ControlNet for forming the graphic to the shape of the clothing item. Next we began implementing 3d clothing meshes and projecting textures onto it, and serving the Stable Diffusion model from a Brev.dev/Google Cloud service. Lastly, we added hand gestures, attached the virtual clothing to the user using BlazePose, and linked the front-end to the back-end and cleaned everything up.

Challenges we ran into

We came to our first major roadblock was when we discovered that the AR body tracking was unable to accurately align 3d clothing to the user's body, but by utilizing some clever math we were able to get a result we were happy with. We also more generally were challenged by the use of new and emerging technologies and their integration.

Accomplishments that we're proud of

Some specific accomplishments we're proud of are our ability to accurately generate and project textures onto any 3d model, to attach virtual clothing in AR, and to use hand gestures to control the ui; not to mention our hard work and dedication to the project.

What we learned

We learned a lot about Stable Diffusion, ControlNet, Three.js, SvelteKit, KinetixML, Brev.dev, Google Cloud, AR, and BlazePose.

What's next for Dreamwear

We would really love to find a more accurate way to align the clothes to the user, add more clothing item models, and improve the hand gesture accuracy.