Fun Things To Do With Selfhosted Generative AI

Weekends are my perfect time to explore, learn new things and experiment with new technologies. For me, one of the most fascinating areas to dive into is generative AI. It allows us to create unique and captivating content by leveraging the power of artificial intelligence.

In this blog post, I’ll share a simple yet exciting thing where I used generative AI with both text and image. Let’s dive in!

Transforming Images with Generative AI

To begin, I utilized a technique called image generation using generative AI. The process involves taking a base image and generating new images based on it. The possibilities are endless, and it’s a great way to explore your creativity. Here’s an example:

This image showcases the incredible potential of generative AI. By starting with a base image, the AI algorithm generates unique variations, each with its own artistic flair. It’s like having an AI-powered art assistant that can help you create visually stunning pieces.

The Workflow

To achieve these impressive results, I followed a specific workflow:

Here’s a breakdown of the steps involved:

1. Load a base image: Choose an image that you want to transform and use as a starting point for the generative AI process.

2. Choose a pre-processing method: Depending on the desired outcome, you can apply various pre-processing techniques to the base image. In this case, I used Depth Map generation to create a kind of initial prototype to direct the AI model when generating new images.

3. Choose a checkpoint: A pre-trained model called DreamShaper was selected for this project. These pre-trained models serve as a foundation for the generative AI process, providing a starting point for generating new images.

4. Instruct the model: To guide the generative AI model, I provided positive and negative prompts. These prompts help the model understand the desired characteristics and style of the final image. By providing clear instructions, you can influence the output and shape the creative process.

5. Enhance the image: After generating the initial image, I applied some post-processing techniques to enhance the quality and refine the details. Techniques like LoRAs (Low-Rank Adaptation) can be used to improve the image’s overall appearance and make it more visually appealing.

Make ChatGPT conversation more enjoyable

I used the newly generated image as the avatar for a chat character in my favourite ChatGPT alternative to create more engaging and interactive experiences:

The chat in this example is powered by Marcoroni-13B, quantized into GGUF format by Tom Jobbins. The model based on Llama-2 and is one of highest scored model in the Open LLM leaderboard. By combining the creativity of generative AI with the conversational abilities of chatbots, we can have more enjoyable and immersive interactions.

Conclusion

Exploring generative AI opens up a world of creative possibilities. By leveraging image generation techniques, we can transform base images into captivating variations. Whether you’re an artist, developer, or simply curious about AI, experimenting with generative AI can be a rewarding and exciting endeavor. So, why not spend your next weekend exploring the limitless potential of generative AI? Let your creativity soar and see where it takes you!