Some Initial Objectives
🎯 Start a small piece of digital art harnessing the power of AI
🎯 Refine my newly re-found programming skills
🎯 Prove open source hashtag#GenerativeAI possibilities
🎯 Take my AI-skills to a next level
🎯 Create an autonomous agent
🎯 Have fun
Defining Overall Structure
Do you wonder how I made all of this happen? There’s no rocket science involved, although the result might launch you straight to Mars 😉.
The system – as shown in the above picture – uses several building blocks to get the job done. It…
- Surfs Google News for the latest headlines.
- Employs an open-source Large Language Model (LLM) to spin a short tale from each title, also imagining a visual description to match.
- Transforms this description into artwork using an image generator.
- Catalogues these creative outputs in a sleek SQLite database.
- Publishes the newly created articles for everyone to read, in this case to a WordPress website.
Et voilà, a setup that marries current events with AI creativity.
Behind the Scenes
Transitioning from a commercial API to SQLite not only saved my wallet but also proved a robust choice for this venture. My LLM of choice after some experimentation with different models, “Mistral Instruct v0.2 7B”, hosted on LMStudio, became my narrative engine. ComfyUI, my open-source stalwart for image generation, played its role to perfection, transforming words into visuals.
Finally, the resulting articles are published on a WordPress website I host myself. Because after all, what’s all the fun if you can’t share the results 😍
This endeavor, albeit only three days old, has already yielded some captivating results, illustrating the untapped potential of open-source tools in standing shoulder-to-shoulder with their commercial counterparts.