Even though I came to know about PartyRock (An Amazon Bedrock Playground) only via Devpost hackathon, I knew exactly what to build after playing with it a bit.
Me and my partner spent countless hours in designing a fusion dress for our wedding to showcase our different cultures, I wanted to create an application which makes the process easier for others and PartyRock was the perfect tool for it.
The intuitive, minimalist nature of PartyRock & blazing fast output helped me to iterate Fusion Dress Designer quickly to create the app I wanted.
Widgets
The widgets in PartyRock are divided into two categories - AI-powered and those which aren't. As of writing, non-AI powered widgets include Static Text and User Input widgets. AI powered widgets include Text Generation, Image Generation and Chatbot.
Static Text
Static text widget can be used to describe the application and to give guidelines for the usage of the application.
User Input
User input widget can be used as a variable for holding the data from the user for further processing using the Large Language Model (LLM). I used it to get the Occasion, Color and Styles for the dress.
I used the placeholder of the user input widget to give suggestions to the user. Default values can be used to help the user generate a sample output during their first use of the application.
Input widgets for Fusion Dress Designer
Text Generation
Text Generation widget creates a LLM generated text using the user inputs. I use it to generate the description for the dress using the occasion, color and styles given by the user.
Image Generation
Image Generation widget generates image based on the given description, I use it to generate the dress image based on the description of the dress generated by the LLM in the previous step.
Chatbot
Chatbot widget allows the users to chat with LLM in the app. This could be useful for fine tuning the results and to avoid clutter of user-input widgets; since there are only 3 user input widgets for Fashion Dress Designer, I'm not using chatbot widget.
Models
PartyRock offers various foundation models like Amazon Titan, Claude, Jurassic, Command, Jurassic, Llama 2 and Stable Diffusion XL as of this writing.
Each model has their strengths and weaknesses; the best model for a application can be found by trial and error with prompt engineering.
I use Jurassic-2 Ultra for dress description generation using LLM as it seems to give a concise text which the text-to-image model (Stable Diffusion XL) could almost always reproduce accurately.
Fusion dress description and dress image
Demo
Bedrock
PartyRock impressive as it is, is a playground for Amazon Bedrock - A fully managed service to build generative AI applications at scale.
Bedrock offers the foundation models seen in the PartyRock and more. It offers model fine-tuning with our own labeled datasets. It has RAG capabilities to supply our data to improve the accuracy and relevancy of the responses. Agents in Bedrock help to manage multi step tasks across several systems and data sources automatically.
Bedrock would be the natural choice for building a real-world Fashion Dress Designer where users can design their own fusion dress and buy them from a boutique.
Agents can help generate better prompts based on user preferences for their dress. Agents can perform tasks related to material inventory, feasibility of design etc.
RAG can improve responses based on data sources from the boutique shop's knowledge base.
Models can be customized and fine-tuned on licensed fashion datasets from leading fashion designers.
Conclusion
Checkout Fashion Dress Designer and create your own fusion dress. Would you like to design your own fusion dress and wear it?
Remix the app to see if you can improve on it and please let me know your feedback!
