📽️ See more video samples in our project page!
"An astronaut floating in space, high quality, 4K resolution."
100 frames, 320 X 512 resolution
"A colony of penguins waddling on an Antarctic ice sheet, 4K, ultra HD."
100 frames, 320 X 512 resolution
[2024.05.25] 🥳🥳🥳 We are thrilled to present our official PyTorch implementation for FIFO-Diffusion. We are releasing the code based on VideoCrafter2.
[2024.05.19] Our paper, FIFO-Diffusion: Generating Infinite Videos from Text without Training, has been archived.
git clone [email protected]:jjihwan/FIFO-Diffusion_public.git
cd FIFO-Diffusion_public
☀️ Start with VideoCrafter
python3 -m venv .fifo
source .fifo/bin/activate
pip install -r requirements.txt
Model | Resolution | Checkpoint |
---|---|---|
VideoCrafter2 (Text2Video) | 320x512 | Hugging Face |
Store them as following structure:
cd FIFO-Diffusion_public
.
└── videocrafter_models
└── base_512_v2
└── model.ckpt # VideoCrafter2 checkpoint
Requires less than 9GB VRAM with Titan XP.
python3 videocrafter_main.py --save_frames
May consume slightly more memory than the single GPU inference (11GB with Titan XP). Please note that our implementation for parallel inference might not be optimal. Pull requests are welcome! 🤓
python3 videocrafter_main_mp.py --num_gpus 8 --save_frames
☀️ Start with Open-Sora Plan (Comming Soon)
cd FIFO-Diffusion_public
git clone [email protected]:PKU-YuanGroup/Open-Sora-Plan.git
python -m venv .sora
source .sora/bin/activate
cd Open-Sora-Plan
pip install -e .
sh scripts/opensora_fifo_ddpm.sh
☀️ Start with zeroscope (Comming Soon)
python3 -m venv .fifo
source .fifo/bin/activate
pip install -r requirements.txt
mkdir zeroscope_models # directory where the model will be stored
python3 zeroscope_main.py
@article{kim2024fifo,
title = {FIFO-Diffusion: Generating Infinite Videos from Text without Training},
author = {Jihwan Kim and Junoh Kang and Jinyoung Choi and Bohyung Han},
journal = {arXiv preprint arXiv:2405.11473},
year = {2024},
}
Our codebase builds on VideoCrafter, Open-Sora Plan, zeroscope. Thanks the authors for sharing their awesome codebases!