Introduction

Welcome to our Akashathon 3 submission! On this page you will find documentation for our project as well as high-level explanations of each component, its features and how they all operate together. For a look at the source code, see our GitHub repo.

Who are we

Dev3 Studio is a small, but dedicated team based in South Africa, passionate about crafting bespoke digital solutions that perfectly align with each client's unique goals. For more information, visit our website.

Our goal

While there are many powerful AI models available online, (Midjourney, Dall-E 3, FLUX.1 [dev]) they rely on complicated parameters like CFG and Samplers which the average user does not understand. Achieving the best results also requires users to make use of prompt optimization, something that users may not be aware of or may not know how to do.

Our goal was to create a simple image generation site that someone with zero knowledge of AI image generation could use while still achieving quality results.

We do this by making use of Llama 3 through the AkashChat API, with its function calling feature to run different image workflows depending on the style of the requested image. We also use Llama to automatically optimize the users prompt which allows users to achieve quality results with very simple prompts. More detail on our approach can be found in the Wrapper API section.

Original prompt: "Anime style photo of astronaut in space, show face in helmet with space in background"

Comparison of general (bottom) vs optimized model (top) and standard (left) vs optimized prompt (right)