🎬 ZeroScope AI Video Generator

Text-to-video generation using the open-source ZeroScope v2 576w diffusion model with a clean web interface. Designed for Colab (GPU recommended) or local installs (demo mode). Convert text prompts into unique AI-generated video clips—no proprietary APIs, just open-source AI!

---

⭐ Features

• ZeroScope v2 576w (Hugging Face): state-of-the-art open-source text-to-video
• Web interface: Easy prompt → video workflow, one-click generation
• Colab compatibility: Zero config, auto public link in Colab
• GPU acceleration: Fast generation on T4/V100/A100 (or slow on CPU)
• CPU fallback/demo mode: Preview the system even without a GPU
• REST API: Easily integrate or automate video generation
• No vendor lock-in: 100% open-source stack, no API keys needed for public models

---

---

🗂️ Why This Structure?

• Colab-first: Anyone can demo with a GPU, no setup or configuration pains.
• Self-contained web & API: All in FastAPI, simple extension/hosting anywhere.
• GPU detection/fallback: Best quality possible for available hardware.
• Minimal external dependencies: Fast, reproducible, open.
• Real reproducible demo: No marketing hype, demo mode for full transparency.

---

🚦 Quick Start (Colab)

1. Open this project in Google Colab:
2. Change runtime to GPU:
   Runtime → Change runtime type → Select T4 GPU
3. Run all cells.
   • Model loads (~3–5min on T4)
   • A public web link is displayed—click to launch the app!
4. Enter a prompt, generate your video, and watch/download the result.

---

💻 System Requirements

Mode	Python	RAM	GPU	Notes
Demo/CPU	3.8+	4GB	None	Slow, low-res video
AI (GPU, best)	3.8+	16GB+	16GB+ VRAM (T4/V100)	Fast, hi-res video

• Minimum: Python 3.8+, 4GB RAM, for demo mode only
• Recommended: Python 3.8+, 16GB+ RAM, NVIDIA GPU (16GB VRAM or more)

---

🛠️ Dependencies

All dependencies are installed automatically in Colab, or run:

pip install fastapi uvicorn diffusers[torch] torch imageio moviepy

---

🔥 Usage Guide

Web Interface

• Open the generated public URL or local app
• Type a descriptive prompt (see built-in examples)
• Click “Generate”
• Wait for video (2–4min on GPU, longer for CPU/demo)
• Watch or download the result!

API (For Developers)

• POST /generate-video
  • JSON: {"prompt": "your text prompt", "duration": 3}
• GET /video/{video_id}
  • Download or stream your generated video

---

📝 Example Prompts

• “A butterfly landing on a sunflower”
• “Colorful paint drops splashing in water”
• “A cute cat playing with yarn”
• “Rain drops falling on green leaves”

---

⚙️ Settings & Configuration

Setting	GPU (Best)	CPU (Demo)
Resolution	576x320	320x256
Frames	16	8–12
Steps	20	10–15
Wait time	2–5min	10–15min

Change these in the code for different speed/quality tradeoffs.

---

🧑🔬 Troubleshooting

• CUDA out of memory:
  Lower num_frames, resolution, or inference steps in config.
• Model not loading:
  Check internet, restart session.
• Web page not loading:
  Wait 30 seconds after startup; check the public/local URL.
• Video takes too long:
  Use a shorter or simpler prompt, run on GPU, or try demo mode.

---

🚧 How It Works (How + Why)

• GPU/CPU detection:
  Loads the best available version of ZeroScope at runtime.
• Optimizations:
  XFormers and attention/vae slicing for efficiency where supported.
• Colab (Threaded server):
  Runs FastAPI server in background, auto-generates preview link.
• REST API + Web UI:
  Clean HTML-only frontend for fast loading and UX, plus well-documented API routes.
• Open-source all the way:
  No paid API, Hugging Face open models, nothing hidden—just reproducible, transparent results.
• Demo mode fallback:
  Lets you see web UI and basic workflow without a GPU.

---