YangKoch AWS

• 

  Jackie Chan & Nini Yang in the film of Zodiac with matching purple outfits

Inspiration: Me and my friends want to look stylish like celebrities we see at concerts, on televised awards, cool events. We want to know where to buy these or similar outfits and how we look in them with or without the celebrity in the generated AI photo. The "shop matching" tool or service, uses traditional haiku (5-7-5 syllable) structure generated from user input to streamline the resulting AI image generation process - helping users visualize their specific needs or shopping preferences.

What it does

The initial functionality is using prompting and LLM to generate images of user's preferred style based on celebrity chosen. An added function is using visual matching technology or the prompts themselves, to locate each clothing component of the outfit online, providing user with the links to buy, and physical location of store if applicable. A future functionality is being able to try on the clothing without physically being present - which we accomplish by having user's personal avatars try on outfits in virtual reality (ex. in the computer screen or spatial technology). This will help users decide if they like the way they look and if they want to buy.

How we built it

We used Party Rock for the backend. Then we used ChatRPG LLM to match text to AI generated content of celebrities and our described avatar. Ex. I type in "Me and Jackie Chan in the film 12 Zodiacs with matching purple outfits" and images of him and myself populate the dashboard which I can select from for my outfit match shopping.

Then we will use some Visual matching API potentially with AI capabilities to scan the AI generated photo and match it to exact/similar available products online.

Challenges we ran into

Getting prompts entered to accurately portray our desired style.

Creating exact scalable replicas of ourselves as a VR avatar using existing technology.

Find the best image scanning technology that accurately matches celebrity clothing items to those available online.

Accomplishments that we're proud of

Combining existing NLP, image, video and generative AI with our new ideas of VR avatars trying on clothing to create an instantaneous and immersive celebrity shopping experience.

What we learned

Prompting to visual AI generation requires extensive training to get desired results.

Creating VR biometric versions of ourselves can be challenging to make sure our body proportions and measurements are accurately reflected when trying on clothes in the VR space.

What's next for YangKoch AWS

We use the app-generated Image and/or Video AI, prompting, or voice to create an avatar of ourselves (using bioscan) in virtual reality. Our avatar then can try on different clothing options in VR and we can see if we like the way it looks on our avatar or not before buying.

Built With

• chatgpt
• chatrpg
• html5
• llm
• partyrock.aws