Those who know me know I’ve worn many hats, and one of them was a TV development hat in Hollywood for years. Priding myself as a rare somebody who can not only code Python, but has also sold TV shows, I decided to just put both hats on at the same time to see what happens. I budgeted $200, three weeks, and a Plus subscription to ChatGPT.
With that, I built a Stochastic Television Engine – a creative slot machine of sorts. Weekly trends go in, new TV show ideas come out. I even have some samples featured on my website if you want a taste. I’ve sprinkled a couple here in this article.
Three weeks + ChatGPT + $200 = 300 show ideas
In an era where technology is reshaping everything – and yes I said everything, please challenge me – the entertainment industry stands on the brink of a revolution. Imagine a world where aspiring creators can generate compelling TV show pitches without the backing of a major studio.
I believe this endeavor marks a significant step towards what we can call the open sourcing of Hollywood.
The Vision: Democratizing Hollywood Development through Open Source AI
This kind of stuff is happening with fancier people than me too. I just saw today that Danny Boyle is shooting his $75M feature, 28 Years Later on an iPhone.
Hollywood has long been the epicenter of entertainment, but its exclusivity has often sidelined innovative ideas from emerging creators. The high costs of production, limited access to industry insiders, and the dominance of established players have created barriers that stifle creativity. My vision was to dismantle these barriers by harnessing the power of generative AI and cloud computing, making the process of creating and sharing TV show ideas accessible to everyone.
Harnessing Generative AI: Crafting Reality TV Show Pitches
Generative AI, particularly models like OpenAI’s GPT-4, has revolutionized content creation by enabling machines to generate human-like text based on given prompts. I tapped into this technology to automate the ideation process. I could have picked any paradigm if TV but I picked “unscripted” first. This is anything non-fiction-ish like reality, reality competition, game shows, factual entertainment, dramatic factual, reality doc – you can invent more subgenres too. There’s no textbook for this, and I’d throw it out if there was because this is a creative industry.
Here’s how it works:
- Data Collection: I curated a dataset of successful reality TV show descriptions, plotlines, and key elements that resonate with audiences.
- Training the Model: Using this dataset, I prompt engineered my way to defining the nuances of crafting engaging show pitches.
- Automating Generation: With initial prompts decided, I developed Python scripts that generate unique and compelling reality TV show ideas, complete with detailed descriptions and potential formats.
The result? An engine capable of producing high-quality show pitches that can spark interest from producers, creators, and even passionate individuals looking to make their mark in the entertainment industry.
Building the Engine: From Code to Content
Creating this engine was a blend of coding, cloud architecture, and strategic app and AI deployment. Here’s a glimpse into the process:
- Environment Setup: I initiated a Python virtual environment to manage dependencies and ensure a clean workspace.
- Scripting the Process: The core of the engine lies in scripts that interact with the AI model, process data, and output structured show pitches.
- Database Integration: I integrated a MySQL database to store generated pitches, enabling easy retrieval and management of content.
- Automating Workflows: Using cron jobs and scheduled scripts, the engine runs autonomously, continuously generating new pitches and updating the website in real-time.
This modular approach not only streamlined the development process but also ensured scalability and maintainability as the project grows.
The Power of the Cloud: Leveraging Google Cloud Platform
Cloud operations were a weak spot for me previously. But in the last year governing AI for my previous employer, I realized a good cloud platform was the only sensible option to work with AI at any scale. I got to know some of the Google Cloud team and I really liked their vibe so that set me on a learning path that led to me acquiring Google Cloud certification.
This helped, and I highly recommend Coursera to study. They have Google-built courses to prepare you for the official Google certification exams. As of writing this, I am halfway through studying for my second GCP certification in Cloud Architecture.
I can’t recommend enough that you build something new on GCP while simultaneously studying it in a Coursera course.
Deploying the engine on a robust cloud infrastructure was crucial for its performance and reliability. GCP provided the ideal environment with its array of services tailored for AI models and scalable applications. Here’s how I utilized GCP:
- Compute Engine: Hosted the backend processes, ensuring that the engine runs smoothly without local hardware constraints. One cheap VM for my actual trend engine, and another as a WordPress server – that you are actually reading this on right now.
- Cloud Storage: Managed and stored large datasets of video and images and generated content efficiently.
- Cloud SQL: Implemented a managed MySQL database, providing high availability and automated backups. I migrated this back to my VM though because it was overkill for cost at this stage in development.
- Google App Engine: Deployed the web interface, allowing users to access and explore the generated show pitches seamlessly.
- Google Vertex: Google Gemini is key to multimodal necessity.
GCP’s flexible pricing models and extensive free tier offerings made it possible to maintain operational costs aligning perfectly with my budget constraints.
Cost Efficiency: Achieving a Functional Stack for Under $200
One of the most compelling aspects of this project is its affordability. Here’s a breakdown of how I kept costs minimal:
- Initial Setup: Leveraged GCP’s free tier services to minimize upfront expenses, keeping the initial cost well below $200.
- Operational Costs: Optimized resource usage by scaling services based on demand, ensuring that daily operational costs remained low.
- Open Source Tools: Utilized free and open-source software for development, further reducing expenses.
- Experimental AI: You can find certain highly expensive multimodal AI models for free in exchange for unpredictability and an utter lack of confidentiality with your prompting. Ideal? Not for real life production, but easily swappable with secure options if you have the budget – and perfect for a prototype.
- Automation: Automated as many processes as possible to eliminate the need for manual intervention, saving both time and money.
- Local AI: Ran Stable Diffusion on a gaming PC I have that already has a great NVIDIA GPU that allowed me to take advantage of CUDA and generate images for free
This cost-effective approach democratizes access to high-powered AI tools and cloud services, making it feasible for individuals and small teams to innovate without hefty financial investments.
Challenges and Triumphs: Navigating the Journey
Embarking on this project wasn’t without its hurdles. But I have to say that the challenges right now that make me realize that AI is not just a hype cycle.
If you are on LinkedIn, you can’t scroll far without seeing either Vermeer’s Girl with the Pearl Earring being used in some AI demo, or the Gartner Hype Cycle being referenced in relation to generative AI.
The challenges and annoyances I hit in three weeks of building this in September of 2024 are precisely why this is not just hype. This is the hardest AI will ever be again.
Let’s hit some of the challenges I faced and how I overcame them – it’s the failures after all that made me smarter in this endeavor:
- Trends: I really wanted to use TikTok, and it was interesting to learn that TikTok does not make it easy to automate around. But most interestingly, what makes TT so relevant as a trend source is precisely this. It moves FAST and has such a high volume of videos that indexing any of it is honestly an commendable accomplishment.
- Integration Complexities: Integrating various services on GCP required a steep learning curve. I dedicated time to understanding GCP’s offerings and best practices, which paid off in the form of a smooth deployment.
- Cost Mistakes: This was the biggest mistake I made. I deployed Flux AI on Vertex to an endpoint for example to generate poster images for the TV show ideas, and that ate about half of my budget immediately. This drive me to setup Sable DIffusion on a local machine that can generate images for free and offload them to a GCP bucket. Less convenient than being houses in the cloud with the rest, but renting those GPUs in cloud gets pricey real quick, and it’s pretty impressive what a good gaming PC can accomplish.
- Scalability: As the engine began generating more pitches, ensuring that the infrastructure could handle the increased load was critical. GCP’s scalable solutions allowed me to adjust resources dynamically without downtime.
- Maintaining Relevance: Keeping the generated pitches fresh and relevant to current trends demanded regular updates to the dataset, a task that required ongoing attention. I have been aiming for fresh inspirational trends every 7 days.
- Bad Web Server: I have a pretty competent background with WordPress, but I decided to challenge myself and try building the site with Ghost because I’ve played with that and kind of like it. It’s free to install on your own server, and I felt like I’d get smarter trying to master it. This just ended up setting me back a few days because no matter what the Ghost server just kept breaking. I redeployed from scratch two or three times over the course of a few days and in the end I just said “f*^k it, this is a waste of time” and I deployed a WP server that you are reading this on now and have had ZERO problems with. Ghost is really cool, but I think it might need a more expert hand than mine to operate on your own server.
- SQL, then SQL, THEN SQL again: With my fresh GCP prowess I tried using BigQuery as a funcitonal database, DESPITE my own studies explaining that a production project like this is the completely wrong case for BQ. I quickly learned why because of data transfer buffers it creates, and this was a big waste of my time because I am already quite proficient with MYSQL that is literally perfect for this project. So I scrapped BQ, and rewrote it to use GCP’s Cloud SQL engine that is amazing – but expensive. I ran this for a day before it ate about 20% of my budget, and immediately transferred it to a perfectly good MYSQL server on the same VM my engine is running – and I just bumped up my VM disk storage by about 20GB. Waaaay cheaper to running an active Cloud SQL instance – that I will ultimately transfer back to with growth and scale as needed – but again, overkill for a $200 prototype.
Despite these challenges, the project’s successes—300 unique reality TV show pitches and a fully operational engine—underscore the potential of combining generative AI with cloud computing.
Implications for the Entertainment Industry
I want to say up front that the pitches and show ideas this engine generates are weird. And this I would say is the point.
The successful deployment of this engine signals a transformative shift in how TV shows can be conceptualized and developed. Here’s what it means for the broader entertainment landscape:
- Increased Accessibility: Aspiring creators no longer need significant financial backing or industry connections to develop and pitch show ideas.
- Diverse Content: With more individuals able to contribute, the range of show concepts becomes more varied, fostering creativity and innovation.
- Rapid Prototyping: Producers can quickly generate and evaluate numerous show ideas, accelerating the development pipeline and reducing time-to-market.
- Democratization of Content Creation: This project exemplifies how technology can level the playing field, allowing talent from all backgrounds to participate in content creation.
Ultimately, this approach could lead to a more inclusive and dynamic entertainment industry, where ideas flourish regardless of their origin.
Future Prospects: What Lies Ahead for Open Source Hollywood
Building this engine is just the beginning. The open sourcing of Hollywood has vast potential, and here’s how I envision its evolution:
- Community Collaboration: Opening up the project to the community can foster collaboration, leading to more sophisticated models and diverse datasets.
- Enhanced AI Capabilities: Integrating more advanced AI features, such as sentiment analysis and trend prediction, can refine the quality and relevance of generated pitches.
- Interactive Platforms: Developing interactive web interfaces where users can customize and generate pitches based on specific genres, themes, or preferences.
- Monetization Opportunities: Exploring avenues like premium content generation, partnerships with production houses, or offering AI-driven consulting services.
- Global Reach: Adapting the engine to cater to different cultures and languages, thereby broadening its applicability and impact.
The journey towards open sourcing Hollywood is ongoing, and I am excited to see how it will reshape the future of entertainment.
Conclusion
The creation of an AI-powered reality TV pitch engine on a modest budget is a testament to the incredible advancements in technology and their potential to democratize industries. By harnessing generative AI and leveraging cost-effective cloud solutions, I’ve taken a significant step towards open sourcing Hollywood—making the creation and sharing of compelling entertainment content accessible to all.
Here’s a favorite of mine from the initial 300. While most of the ideas are of questionable quality – that’s the point – good ideas don’t come easy. We’ve always needed to seek the diamonds in the rough.
💡
Logline
Explore the heartwarming stories of pets who accompany their owners during emergencies, showcasing the unbreakable bond between humans and their furry companions.
As technology continues to evolve, so too will the opportunities for innovation in entertainment. I invite fellow creators, technologists, and enthusiasts to join me in this journey, pushing the boundaries of what’s possible and reshaping the landscape of Hollywood for generations to come.
Feel free to reach out with questions, suggestions, or collaboration ideas. Together, we can revolutionize the entertainment industry.
Leave a Reply