Alice-T2V-14B-MoE

A state-of-the-art text-to-video generation model built on a Mixture-of-Experts (MoE) architecture.

Overview

Alice-T2V-14B-MoE is a large-scale video generation model capable of producing high-quality videos from text prompts. The model supports both 480P and 720P resolutions with 5-second video generation.

Key Features

• MoE Architecture: Utilizes specialized expert models for different stages of the denoising process, maximizing generation quality while maintaining computational efficiency
• Multi-Resolution Support: Generate videos at 480P or 720P resolution
• High-Quality Motion: Trained on extensive video data for realistic and coherent motion generation

Model Details

Specification	Value
Architecture	Mixture-of-Experts (MoE)
Total Parameters	~27B
Active Parameters	~14B per step
Output Resolution	480P, 720P
Video Length	5 seconds

Installation

1. Install dependencies

pip install torch>=2.4.0 diffusers safetensors

2. Install Alice model class

# From the model directory
cd alice
pip install .

Or install directly from the repo:

pip install git+https://huggingface.co/gomirageai/Alice-T2V-14B-MoE.git#subdirectory=alice

Usage

from alice import AliceModel

# Load high-noise expert
high_noise_model = AliceModel.from_pretrained(
    "gomirageai/Alice-T2V-14B-MoE",
    subfolder="high_noise_model"
)

# Load low-noise expert
low_noise_model = AliceModel.from_pretrained(
    "gomirageai/Alice-T2V-14B-MoE",
    subfolder="low_noise_model"
)

License

This model is released under the Apache 2.0 License. You are free to use the model and any content you generate, provided your usage complies with applicable laws and does not cause harm.

Disclaimer

Users are fully responsible for their use of this model. Generated content must not violate laws, harm individuals, spread misinformation, or target vulnerable populations.