Building Apps with the Llama 8B Model

Hey everyone! we recently finished building a fun little project, and I thought it’d be great to share it with others who love experimenting with generative AI. It’s a prototype app that helps you auto-generate prompts for Midjourney. Think of it as a simple way to spark creativity without overthinking the details.

Here’s a quick rundown of how I built it and what I discovered during the process!

The Problem: Prompting Takes Time

When I first started playing around with Midjourney, I realized that creating prompts could sometimes be time-consuming—especially if you’re experimenting with different styles. I also noticed that some people I know (friends, colleagues, or even myself on a tired day) felt stuck when it came to coming up with the "right" words for prompts.

So, I started wondering:

  • Could I make generating prompts faster?
  • Could I make it easier for people who are new to Midjourney?

 

These questions became the foundation for my project.

The Building Process

To build the app, I focused on two goals:

  1. Make it simple for users to input an idea (like “dog”) and quickly get useful prompts.
  2. Ensure the app is fast and accessible, even for beginners.

 

I explored different AI models, testing Llama 3.1 in both the 8B and 405B versions. While the 405B model was impressive in terms of complexity, I ultimately chose the 8B model because of its speed. It could generate prompts in under within seconds, making it perfect for a tool aimed at quick experimentation (Generation speed may vary depending on server load). For example, the app could generate:


To quickly bring this idea to life, I used Gradio to build a simple prototype for testing the app. Gradio made it easy to create a clean, interactive interface where users could type a prompt, select an aspect ratio, and instantly see the results. Its user-friendly setup saved me time and allowed me to focus on the app’s functionality rather than getting bogged down in complex development tools.

The app works like this:

  1. You type a simple idea, like “cat” or “dog.”
  2. Choose an aspect ratio: square, vertical, or horizontal.
  3. Instantly, you get a prompt ready to use in Midjourney!

Addressing Two Main Problems

While building the app, we identified two key problems that marketers and designers often face:

  1. Repetitive Adjustments for Different Ratios: Creators often need to design for multiple platforms, each with different aspect ratio requirements. For example, a YouTube banner should be horizontal, IG reels need to be vertical, and IG posts are square. Manually adding these parameters to Midjourney prompts can be repetitive (and sometimes it’s easy to forget to add the aspect ratio parameter), making the process time-consuming.
  2. Lack of Prompt Ideas: Sometimes, creators struggle with a lack of ideas to kickstart their creative process. They need inspiration or a way to streamline generating effective prompts without overthinking. That’s where the idea of an automated app came in—to handle this process and provide a creative boost.

 

To address these challenges, we built this app based on an open-source model, making it accessible and flexible for a variety of use cases.

Testing and Feedback

After building the prototype, I shared it with some friends—designers, marketers, and even a few people with no Midjourney experience. Here’s what I learned:

  • For Designers and Marketers: They found that the app boosted their workflow, increasing their output by about 50%. The ability to quickly generate prompts meant they could try more styles and refine their visuals faster.
  • For Newbies: Even those with no experience in Midjourney found it easy to use. They could type a single word like “dog” and immediately get a prompt that worked. One user said, “I never thought I’d feel confident using AI tools—this makes it so approachable!”

 

What Building This Taught Me

  • Choosing the Right Model Matters: Bigger isn’t always better when it comes to building apps with AI. The key is to pick a model that accurately meets user needs and delivers the expected results efficiently. For this project, the 8B model struck the perfect balance between performance and speed.
  • Simplicity is Key: People appreciate tools that are straightforward and don’t overwhelm them with too many options.
  • AI is Fun to Build With: This project reminded me why I love working with generative AI—it’s all about making creativity accessible and fun.

 

Conclusion

This project was a small but meaningful step in exploring what generative AI can offer. Building tools like this allows me to combine my passion for technology with creativity, and it’s incredibly rewarding to see how AI can simplify workflows and unlock new possibilities for others. I wanted to share this experience not just to showcase the app, but to connect with others who share the same curiosity and excitement about AI.

Happy building and experimenting!

Back to blog