Spaces:
Running
Running
| title: ClipScript | |
| emoji: '🎬' | |
| colorFrom: pink | |
| colorTo: gray | |
| sdk: gradio | |
| sdk_version: 5.33.1 | |
| app_file: app.py | |
| pinned: false | |
| license: mit | |
| short_description: Transforms videos and audio into ready-to-publish blogs. | |
| tags: | |
| - agent-demo-track | |
| video_overview: https://youtu.be/8DUxlj79NqM | |
| # 🎬 ClipScript: Video-to-Blog Transformer | |
| ClipScript is a powerful application that transforms any video or audio content into a polished, ready-to-publish blog post. Simply provide a YouTube URL or upload an audio file, and let our AI agent handle the rest. | |
| ### Video Overview | |
| [Watch a video demonstrating how to use ClipScript and what it is abut here!](https://youtu.be/8DUxlj79NqM) | |
| ## Features | |
| - **YouTube & File Uploads**: Works with YouTube links or direct audio/video file uploads. | |
| - **AI-Powered Transcription**: Utilizes a state-of-the-art ASR model for highly accurate transcription. | |
| - **Agentic Blog Generation**: An expert AI writing agent converts the raw transcript into a structured, engaging blog post, automatically removing conversational filler and adding SEO-friendly formatting. | |
| - **Interactive Refinement**: Chat with the AI agent to refine the generated blog post until it's perfect. | |
| - **Secure & Scalable**: Powered by [Modal](https://modal.com) for secure, scalable, and efficient backend processing. | |
| ## Hugging Face Agent Demo Track | |
| This application has been submitted to the **Agent Demo Track**. It showcases an "AI agent" that acts as an expert blog writer and editor, taking a high-level goal (transforming a transcript) and executing a series of steps to achieve it. | |
| ## Core Technology | |
| ### Speech-to-Text: NVIDIA Parakeet TDT 0.6B V2 | |
| The transcription engine is powered by `nvidia/parakeet-tdt-0.6b-v2`. This model is **ranked #1 on the Hugging Face Open ASR Leaderboard**, achieving the best overall average Word Error Rate (WER) and RTFx (real-time factor) score, making it one of the fastest and most accurate ASR models available. | |
| For a deep dive into the model's architecture and performance, check out the [official model card](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2) and the [Open ASR Leaderboard](https://huggingface.co/spaces/hf-audio/open_asr_leaderboard). | |
| For audio longer than 30 minutes, the SST model automatically segments content into optimal chunks and processes them in parallel, enabling fast transcription of hours-long content while maintaining accuracy and context. | |
| ### Content Generation: AI Writing Agent | |
| An AI writing agent, accessed via OpenRouter, converts the raw transcript into a polished, structured blog post, ready for publishing. | |
| ### Backend Infrastructure: Modal | |
| The backend is built on [Modal](https://modal.com) for security, scalability, and performance. | |
| - **Secure Sandboxed Execution**: All media processing occurs in isolated Modal environments, keeping potentially malicious files separate from the Gradio server. | |
| - **High-Performance File System**: Modal Volumes provide fast, reliable file transfer and access for user uploads. | |
| This architecture keeps the frontend lightweight while offloading intensive tasks to secure, scalable cloud resources. | |
| ## Architecture | |
| The following diagram illustrates the complete data flow, from user input in the Gradio application to the final blog post generation. | |
|  | |
| Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference | |