metadata

pipeline_tag: text-to-video
license: cc-by-nc-4.0

Info

base model zeroscope_v2_576w.
Train on potat1 dataset with 50000 steps for 10 hours on single GPU of 24GB VRAM.
Thought result of training is not good.

Install Windows

git clone https://github.com/ExponentialML/Text-To-Video-Finetuning.git
cd Text-To-Video-Finetuning
git lfs install
git clone https://huggingface.co/damo-vilab/text-to-video-ms-1.7b ./models/model_scope_diffusers/
py -m venv --clear venv && venv\Scripts\activate
pip install -r requirements.txt
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118 --no-cache --force-reinstall --version --isolated --ignore-installed

git clone https://github.com/ExponentialML/Video-BLIP2-Preprocessor.git
cd Video-BLIP2-Preprocessor
pip install -r requirements.txt

:: fix1
echo accelerate^>=0.20.3>>requirements-dev.txt
pip install -r requirements-dev.txt

:: Captation
python preprocess.py --video_directory C:\Video-BLIP2-Preprocessor\videos --config_name "My Videos" --config_save_name "my_videos"

:: Training
venv\Scripts\activate && python train.py --config ./configs/v2/train_config.yaml

:: inference  1024x576
python inference.py --model zeroscope_v2_576w_potat1\zeroscope_v2_576w-checkpoint-50000 --prompt "a fast moving fancy sports car" --fps 24 --num-frames 30 --window-size 12 --width 1024 --height 576 --sdp