Inspiration
In a world increasingly influenced by social media trends and influencers, we’ve noticed that developing new, engaging trends is becoming more difficult. This challenge spans all categories of short videos. We believe that using Generative AI (GenAI) to ideate new trends based on inspiration from existing ones can significantly diversify content and better engage audiences.
What it does
Our app analyzes trending TikTok videos and generates new video ideas in the form of short descriptions, along with novel AI-generated music tracks to serve as background scores and image generation features, providing visual templates to further enhance the creative process. This combination helps creators develop new and engaging content.
How we built it
d
🌍 Market Impact
TrendTok is set to transform the creative industry! By harnessing AI's endless creativity, creators can keep their content fresh and engaging, leading to endless entertainment for users and maximized engagement for creators. This innovation enhances content diversity and richness on TikTok, driving higher user engagement and retention. The increased activity benefits creators through greater exposure and monetization opportunities, while solidifying TikTok's position as a leading social media platform. TrendTok’s approach sets new standards for creativity and user interaction, leading to greater advertising revenue and robust community interactions. By revolutionizing content creation and consumption, TrendTok is shaping the future of digital entertainment.
Challenges we ran into
- Data Access: Scraping TikTok data required access to APIs only available to doctoral researchers, so we explored alternative methods. We used web scraping, third-party APIs, social media monitoring tools, public data repositories, data donations from TikTok users, and manual data collection.
- Model Availability: Obtaining open-source GenAI models was challenging due to the limited availability of music & image generator models and free LLMs suitable for cloud-based API calls.
- Compute Power: The high compute power required for video and image generation models necessitated using lighter models and cloud-based APIs.
Accomplishments that we're proud of
We are incredibly proud of the groundbreaking achievements we accomplished with our project:
Revolutionizing Trend Generation: We have pioneered a sophisticated AI-driven pipeline that generates not only textual ideas but also original audio tracks and image visualization templates, enabling creators to develop unique and engaging content that stands out in the crowded TikTok landscape.
Cutting-Edge AI Integration: Our application seamlessly integrates state-of-the-art AI technologies, including Llama 3 LLM for idea generation, Meta MusicGen for music synthesis and StableDiffusion for image visualization templates, showcasing our ability to harness advanced models for practical, creative solutions.
Enhanced User Engagement: By providing influencers with ready-to-use templates for new trends, we empower them to captivate their audiences with fresh, innovative content, thus driving higher levels of engagement and interaction on the platform.
Robust Data Processing: We successfully implemented a comprehensive data collection and processing framework, utilizing web scraping and NLP techniques to extract and distill relevant information from trending videos, ensuring our AI models have the best possible data to work with.
Scalable and Secure Infrastructure: Our solution is deployed in a Docker container, providing a degree of encapsulation and security for identifiable data which is sent over the internet. We also intend to deploy it in a Kubernetes instance to scale our app for use by a much bigger audience.
🧠 What We Learned
We learned to deploy cutting-edge Generative AI models in text, video, audio, and image generation. We experienced firsthand the challenges of the compute power needed for inferencing large models for image and video generation. Additionally, we gained valuable insights into integrating various technologies seamlessly to create a cohesive and functional product.
🚀 What's Next
We plan to test our model on historical trend data to measure its accuracy in predicting future trends, thereby demonstrating the pipeline's robustness. Given the extended inference times for large models on PCs, we will utilize a dedicated cloud machine for these tasks. Due to limitations in collecting real-time data from TikTok, we plan to integrate this feature in the future by obtaining the necessary permissions. Furthermore, we plan to explore partnerships with content creators and influencers to refine our application and expand its reach within the TikTok community.
Log in or sign up for Devpost to join the conversation.