LEGION-Video-Gen

---

license: apache-2.0 library_name: diffusers tags:

• text-to-video
• image-to-video
• video-generation
• diffusers pipeline_tag: text-to-video inference: true base_model: deathlegionteam/LEGION-Video-Gen widget:
• text: "A serene mountain lake at sunset with colorful clouds reflecting on the water"

⚔️ LEGION VIDEO GENERATION — The Ultimate AI Video Engine

<p align="center"> <strong>State-of-the-art video generation with 8.3B parameters</strong><br> Text-to-Video · Image-to-Video · QWatermark System </p> <p align="center"> <img src="https://img.shields.io/badge/Params-8.3B-blue" alt="Parameters"> <img src="https://img.shields.io/badge/License-Apache%202.0-green" alt="License"> <img src="https://img.shields.io/badge/GPU-Recommended-red" alt="GPU"> <a href="https://huggingface.co/deathlegionteam/LEGION-Video-Gen"><img src="https://img.shields.io/badge/🤗%20HuggingFace-LEGION--Video--Gen-blue" alt="HuggingFace"></a> </p>

📋 Table of Contents

• ✨ Features
• 🚀 Quick Start
• 🌐 API Documentation
• 💧 QWatermark System
• 🤗 HuggingFace
• 🖥️ Project Structure
• 🎬 Example Prompts
• 📜 License

✨ Features

• 🎬 Text-to-Video Generation — Create videos from any text prompt with cinematic quality
• 🖼️ Image-to-Video Generation — Animate still images with controlled motion
• 💧 QWatermark System — Configurable semi-transparent quality assurance watermark with position, size, opacity, and text controls
• 🌐 Web Application — Full Gradio UI with dark theme and FastAPI backend
• 📡 REST API — Programmatic video generation via HTTP endpoints
• 🛡️ Graceful Fallback — Mock generation mode when no GPU is available

🚀 Quick Start

Prerequisites

• GPU (Recommended): NVIDIA GPU with 16GB+ VRAM (RTX 4090, A100, H100)
• CPU (Fallback): Works with mock generation mode (test pattern videos)
• Python 3.10+
• ~30GB free disk space (model weights)

Installation

# Clone the repository
git clone https://huggingface.co/deathlegionteam/LEGION-Video-Gen
cd LEGION-Video-Gen

# Create virtual environment
python3 -m venv venv
source venv/bin/activate

# Install dependencies
pip install --upgrade pip
pip install -r requirements.txt

# Verify installation
python3 -c "import torch, diffusers, gradio, fastapi; print('OK')"

Quick Start — Generate Your First Video

from inference import LegionVideoGenerator

generator = LegionVideoGenerator()
video_path = generator.generate_from_text(
    prompt="A serene mountain lake at sunset with colorful clouds reflecting on the water, gentle ripples, cinematic quality",
    num_frames=49,
    width=480,
    height=480,
    num_inference_steps=50,
    guidance_scale=6.0,
    watermark_strength=0.3,
)
print(f"Video saved to: {video_path}")

Starting the Web UI

# Start the API backend
python3 backend/main.py &

# Start the Gradio frontend
python3 frontend/app.py

# Open http://localhost:8080 in your browser

🌐 API Documentation

REST API Endpoints

The backend runs on port 8081 by default.

Text-to-Video Generation

import requests

response = requests.post(
    "http://localhost:8081/api/generate/text",
    json={
        "prompt": "A cyberpunk city street at night with neon lights reflecting on wet pavement",
        "negative_prompt": "warped, distorted, flickering, jittery, low quality, blurry, artifacts",
        "num_frames": 49,
        "width": 480,
        "height": 480,
        "num_inference_steps": 50,
        "guidance_scale": 6.0,
        "watermark_strength": 0.3,
    }
)

with open("output.mp4", "wb") as f:
    f.write(response.content)

Image-to-Video Generation

import requests

with open("input_image.jpg", "rb") as img:
    response = requests.post(
        "http://localhost:8081/api/generate/image",
        files={"file": img},
        data={
            "prompt": "Gentle motion, cinematic camera movement, atmospheric",
            "num_frames": 49,
            "width": 480,
            "height": 480,
            "num_inference_steps": 50,
            "guidance_scale": 6.0,
            "watermark_strength": 0.3,
        }
    )

with open("animated.mp4", "wb") as f:
    f.write(response.content)

💧 QWatermark System

The QWatermark (Quality Watermark) system imprints a configurable assurance marker on every generated video.

🤗 HuggingFace

• Model Repository: deathlegionteam/LEGION-Video-Gen
• Space (Live Demo): deathlegionteam/LEGION-Video-Gen-Space

Model Weights

The model is available as a complete Diffusers pipeline on HuggingFace Hub. You can load it directly using the Diffusers library:

from diffusers import DiffusionPipeline
import torch

pipe = DiffusionPipeline.from_pretrained(
    "deathlegionteam/LEGION-Video-Gen",
    torch_dtype=torch.float16,
)
pipe = pipe.to("cuda")
pipe.vae.enable_tiling()
pipe.enable_attention_slicing()

# Generate video
video_frames = pipe(
    prompt="A serene mountain lake at sunset",
    num_frames=49,
    width=480,
    height=480,
    num_inference_steps=50,
    guidance_scale=6.0,
).frames[0]

🖥️ Project Structure

/app/video_generation_pipeline_1006/
├── inference.py           # Core generation class (LegionVideoGenerator)
├── backend/
│   └── main.py            # FastAPI backend (port 8081)
├── frontend/
│   ├── app.py             # Gradio frontend (port 8080)
│   └── streamlit_app.py   # Streamlit frontend
├── models/
│   ├── t2v/               # T2V model weights (safetensor format)
│   └── i2v/               # I2V model directory
├── outputs/               # Generated videos
├── requirements.txt       # Python dependencies
├── README.md              # This file
└── .space/                # HuggingFace Space configuration

🎬 Example Prompts

Text-to-Video

Image-to-Video

📊 Performance

📝 Notes

• GPU Required for Real Inference: The 8.3B parameter model requires ~16GB VRAM for FP16 inference. Without a GPU, the system runs in mock mode.
• Disk Space: Full model weights (T2V) are approximately 13GB. Additional I2V variant would add another ~13GB.

📜 License

This project is licensed under Apache 2.0.

<p align="center"> <strong>⚔️ LEGION VIDEO GENERATION</strong><br> Built with ❤️ for the open-source AI community </p>