Home

Awesome

CVPR 2024 | PIA：Personalized Image Animator

PIA: Your Personalized Image Animator via Plug-and-Play Modules in Text-to-Image Models

Yiming Zhang*, Zhening Xing*, Yanhong Zeng†, Youqing Fang, Kai Chen†

(*equal contribution, †corresponding Author)

arXiv Project Page Open in OpenXLab Third Party Colab HuggingFace Model <a target="_blank" href="https://huggingface.co/spaces/Leoxing/PIA"> <img src="https://huggingface.co/datasets/huggingface/badges/raw/main/open-in-hf-spaces-sm.svg" alt="Open in HugginFace"/> </a> Replicate

PIA is a personalized image animation method which can generate videos with high motion controllability and strong text and image alignment.

If you find our project helpful, please give it a star :star: or cite it, we would be very grateful :sparkling_heart: .

<img src="__assets__/image_animation/teaser/teaser.gif">

What's New

Setup

Prepare Environment

Use the following command to install a conda environment for PIA from scratch:

conda env create -f pia.yml
conda activate pia

You may also want to install it based on an existing environment, then you can use environment-pt2.yaml for Pytorch==2.0.0. If you want to use lower version of Pytorch (e.g. 1.13.1), you can use the following command:

conda env create -f environment.yaml
conda activate pia

We strongly recommend you to use Pytorch==2.0.0 which supports scaled_dot_product_attention for memory-efficient image animation.

Download checkpoints

<li>Download the Stable Diffusion v1-5</li> 
conda install git-lfs
git lfs install
git clone https://huggingface.co/runwayml/stable-diffusion-v1-5 models/StableDiffusion/
<li>Download PIA</li> 
git clone https://huggingface.co/Leoxing/PIA models/PIA/
<li>Download Personalized Models</li> 
bash download_bashscripts/1-RealisticVision.sh
bash download_bashscripts/2-RcnzCartoon.sh
bash download_bashscripts/3-MajicMix.sh

You can also download pia.ckpt manually through link on Google Drive or HuggingFace.

Put checkpoints as follows:

└── models
    ├── DreamBooth_LoRA
    │   ├── ...
    ├── PIA
    │   ├── pia.ckpt
    └── StableDiffusion
        ├── vae
        ├── unet
        └── ...

Inference

Image Animation

Image to Video result can be obtained by:

python inference.py --config=example/config/lighthouse.yaml
python inference.py --config=example/config/harry.yaml
python inference.py --config=example/config/majic_girl.yaml

Run the command above, then you can find the results in example/result:

<table class="center"> <tr> <td><p style="text-align: center">Input Image</p></td> <td><p style="text-align: center">lightning, lighthouse</p></td> <td><p style="text-align: center">sun rising, lighthouse</p></td> <td><p style="text-align: center">fireworks, lighthouse</p></td> </tr> <tr> <td><img src="example/img/lighthouse.jpg"></td> <td><img src="__assets__/image_animation/real/1.gif"></td> <td><img src="__assets__/image_animation/real/2.gif"></td> <td><img src="__assets__/image_animation/real/3.gif"></td> </tr> <tr> <td><p style="text-align: center">Input Image</p></td> <td><p style="text-align: center">1boy smiling</p></td> <td><p style="text-align: center">1boy playing the magic fire</p></td> <td><p style="text-align: center">1boy is waving hands</p></td> </tr> <tr> <td><img src="example/img/harry.png"></td> <td><img src="__assets__/image_animation/rcnz/1.gif"></td> <td><img src="__assets__/image_animation/rcnz/2.gif"></td> <td><img src="__assets__/image_animation/rcnz/3.gif"></td> </tr> <tr> <td><p style="text-align: center">Input Image</p></td> <td><p style="text-align: center">1girl is smiling</p></td> <td><p style="text-align: center">1girl is crying</p></td> <td><p style="text-align: center">1girl, snowing </p></td> </tr> <tr> <td><img src="example/img/majic_girl.jpg"></td> <td><img src="__assets__/image_animation/majic/1.gif"></td> <td><img src="__assets__/image_animation/majic/2.gif"></td> <td><img src="__assets__/image_animation/majic/3.gif"></td> </tr> </table> <!-- More results: <table class="center"> <tr> <td><p style="text-align: center">Input Image</p></td> </tr> <tr> </tr> </table> -->

Motion Magnitude

You can control the motion magnitude through the parameter magnitude:

python inference.py --config=example/config/xxx.yaml --magnitude=0 # Small Motion
python inference.py --config=example/config/xxx.yaml --magnitude=1 # Moderate Motion
python inference.py --config=example/config/xxx.yaml --magnitude=2 # Large Motion

Examples:

python inference.py --config=example/config/labrador.yaml
python inference.py --config=example/config/bear.yaml
python inference.py --config=example/config/genshin.yaml
<table class="center"> <tr> <td><p style="text-align: center">Input Image<br>& Prompt</p></td> <td><p style="text-align: center">Small Motion</p></td> <td><p style="text-align: center">Moderate Motion</p></td> <td><p style="text-align: center">Large Motion</p></td> </tr> <tr> <td><img src="example/img/labrador.png" style="width: 220px">a golden labrador is running</td> <td><img src="__assets__/image_animation/magnitude/labrador/1.gif"></td> <td><img src="__assets__/image_animation/magnitude/labrador/2.gif"></td> <td><img src="__assets__/image_animation/magnitude/labrador/3.gif"></td> </tr> <tr> <td><img src="example/img/bear.jpg" style="width: 220px">1bear is walking, ...</td> <td><img src="__assets__/image_animation/magnitude/bear/1.gif"></td> <td><img src="__assets__/image_animation/magnitude/bear/2.gif"></td> <td><img src="__assets__/image_animation/magnitude/bear/3.gif"></td> </tr> <tr> <td><img src="example/img/genshin.jpg" style="width: 220px">cherry blossom, ...</td> <td><img src="__assets__/image_animation/magnitude/genshin/1.gif"></td> <td><img src="__assets__/image_animation/magnitude/genshin/2.gif"></td> <td><img src="__assets__/image_animation/magnitude/genshin/3.gif"></td> </tr> </table>

Style Transfer

To achieve style transfer, you can run the command(Please don't forget set the base model in xxx.yaml):

Examples:

python inference.py --config example/config/concert.yaml --style_transfer
python inference.py --config example/config/anya.yaml --style_transfer
<table class="center"> <tr> <td><p style="text-align: center">Input Image<br> & Base Model</p></td> <td><p style="text-align: center">1man is smiling</p></td> <td><p style="text-align: center">1man is crying</p></td> <td><p style="text-align: center">1man is singing</p></td> </tr> <tr> <td style="text-align: center"><img src="example/img/concert.png" style="width:220px">Realistic Vision</td> <td><img src="__assets__/image_animation/style_transfer/concert/1.gif"></td> <td><img src="__assets__/image_animation/style_transfer/concert/2.gif"></td> <td><img src="__assets__/image_animation/style_transfer/concert/3.gif"></td> </tr> <tr> <td style="text-align: center"><img src="example/img/concert.png" style="width:220px">RCNZ Cartoon 3d</td> <td><img src="__assets__/image_animation/style_transfer/concert/4.gif"></td> <td><img src="__assets__/image_animation/style_transfer/concert/5.gif"></td> <td><img src="__assets__/image_animation/style_transfer/concert/6.gif"></td> </tr> <tr> <td><p style="text-align: center"></p></td> <td><p style="text-align: center">1girl smiling</p></td> <td><p style="text-align: center">1girl open mouth</p></td> <td><p style="text-align: center">1girl is crying, pout</p></td> </tr> <tr> <td style="text-align: center"><img src="example/img/anya.jpg" style="width:220px">RCNZ Cartoon 3d</td> <td><img src="__assets__/image_animation/style_transfer/anya/1.gif"></td> <td><img src="__assets__/image_animation/style_transfer/anya/2.gif"></td> <td><img src="__assets__/image_animation/style_transfer/anya/3.gif"></td> </tr> </table>

Loop Video

You can generate loop by using the parameter --loop

python inference.py --config=example/config/xxx.yaml --loop

Examples:

python inference.py --config=example/config/lighthouse.yaml --loop
python inference.py --config=example/config/labrador.yaml --loop
<table> <tr> <td><p style="text-align: center">Input Image</p></td> <td><p style="text-align: center">lightning, lighthouse</p></td> <td><p style="text-align: center">sun rising, lighthouse</p></td> <td><p style="text-align: center">fireworks, lighthouse</p></td> </tr> <tr> <td style="text-align: center"><img src="example/img/lighthouse.jpg" style="width:auto"></td> <td><img src="__assets__/image_animation/loop/lighthouse/1.gif"></td> <td><img src="__assets__/image_animation/loop/lighthouse/2.gif"></td> <td><img src="__assets__/image_animation/loop/lighthouse/3.gif"></td> </tr> <tr> <td><p style="text-align: center">Input Image</p></td> <td><p style="text-align: center">labrador jumping</p></td> <td><p style="text-align: center">labrador walking</p></td> <td><p style="text-align: center">labrador running</p></td> </tr> <tr> <td style="text-align: center"><img src="example/img/labrador.png" style="width:auto"></td> <td><img src="__assets__/image_animation/loop/labrador/1.gif"></td> <td><img src="__assets__/image_animation/loop/labrador/2.gif"></td> <td><img src="__assets__/image_animation/loop/labrador/3.gif"></td> </tr> </table>

Training

We provide training script for PIA. It borrows from AnimateDiff heavily, so please prepare the dataset and configuration files according to the guideline.

After preparation, you can train the model by running the following command using torchrun:

torchrun --nnodes=1 --nproc_per_node=1 train.py --config example/config/train.yaml

or by slurm,

srun --quotatype=reserved --job-name=pia --gres=gpu:8 --ntasks-per-node=8 --ntasks=8  --cpus-per-task=4 --kill-on-bad-exit=1 python train.py --config example/config/train.yaml

AnimateBench

We have open-sourced AnimateBench on HuggingFace which includes images, prompts and configs to evaluate PIA and other image animation methods.

BibTex

@inproceedings{zhang2024pia,
  title={Pia: Your personalized image animator via plug-and-play modules in text-to-image models},
  author={Zhang, Yiming and Xing, Zhening and Zeng, Yanhong and Fang, Youqing and Chen, Kai},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={7747--7756},
  year={2024}
}

Contact Us

Yiming Zhang: zhangyiming@pjlab.org.cn

Zhening Xing: xingzhening@pjlab.org.cn

Yanhong Zeng: zengyh1900@gmail.com

Acknowledgements

The code is built upon AnimateDiff, Tune-a-Video and PySceneDetect

You may also want to try other project from our team: <a target="_blank" href="https://github.com/open-mmlab/mmagic"> <img src="https://github.com/open-mmlab/mmagic/assets/28132635/15aab910-f5c4-4b76-af9d-fe8eead1d930" height=20 alt="MMagic"/> </a>