[Paper] โ [Project Page] โ [Model Card]
[๐ค Demo (Realistic)] โ [๐ค Demo (Stylization)]
[Replicate Demo (Realistic)] โ [Replicate Demo (Stylization)]
If the ID fidelity is not enough for you, please try our stylization application, you may be pleasantly surprised.
Official implementation of PhotoMaker: Customizing Realistic Human Photos via Stacked ID Embedding.
- Rapid customization within seconds, with no additional LoRA training.
- Ensures impressive ID fidelity, offering diversity, promising text controllability, and high-quality generation.
- Can serve as an Adapter to collaborate with other Base Models alongside LoRA modules in community.
โโ Note: If there are any PhotoMaker based resources and applications, please leave them in the discussion and we will list them in the Related Resources section in README file. Now we know the implementation of Replicate, Windows, ComfyUI, and WebUI. Thank you all!
- โ
Jan. 20, 2024. An important note: For those GPUs that do not support bfloat16, please change this line to
torch_dtype = torch.float16
, the speed will be greatly improved (1min/img (before) vs. 14s/img (after) on V100). The minimum GPU memory requirement for PhotoMaker is 15G. - โ Jan. 15, 2024. We release PhotoMaker.
Note: only change the base model and add the LoRA modules for better stylization
- Python >= 3.8 (Recommend to use Anaconda or Miniconda)
- PyTorch >= 2.0.0
conda create --name photomaker python=3.10
conda activate photomaker
pip install -U pip
# Install requirements
pip install -r requirements.txt
# Install photomaker
pip install git+https://github.com/TencentARC/PhotoMaker.git
Then you can run the following command to use it
from photomaker import PhotoMakerStableDiffusionXLPipeline
The model will be automatically downloaded through the following two lines:
from huggingface_hub import hf_hub_download
photomaker_path = hf_hub_download(repo_id="TencentARC/PhotoMaker", filename="photomaker-v1.bin", repo_type="model")
You can also choose to download manually from this url.
Use like diffusers
- Dependency
import torch
import os
from diffusers.utils import load_image
from diffusers import EulerDiscreteScheduler
from photomaker import PhotoMakerStableDiffusionXLPipeline
### Load base model
pipe = PhotoMakerStableDiffusionXLPipeline.from_pretrained(
base_model_path, # can change to any base model based on SDXL
torch_dtype=torch.bfloat16,
use_safetensors=True,
variant="fp16"
).to(device)
### Load PhotoMaker checkpoint
pipe.load_photomaker_adapter(
os.path.dirname(photomaker_path),
subfolder="",
weight_name=os.path.basename(photomaker_path),
trigger_word="img" # define the trigger word
)
pipe.scheduler = EulerDiscreteScheduler.from_config(pipe.scheduler.config)
### Also can cooperate with other LoRA modules
# pipe.load_lora_weights(os.path.dirname(lora_path), weight_name=lora_model_name, adapter_name="xl_more_art-full")
# pipe.set_adapters(["photomaker", "xl_more_art-full"], adapter_weights=[1.0, 0.5])
pipe.fuse_lora()
- Input ID Images
### define the input ID images
input_folder_name = './examples/newton_man'
image_basename_list = os.listdir(input_folder_name)
image_path_list = sorted([os.path.join(input_folder_name, basename) for basename in image_basename_list])
input_id_images = []
for image_path in image_path_list:
input_id_images.append(load_image(image_path))
- Generation
# Note that the trigger word `img` must follow the class word for personalization
prompt = "a half-body portrait of a man img wearing the sunglasses in Iron man suit, best quality"
negative_prompt = "(asymmetry, worst quality, low quality, illustration, 3d, 2d, painting, cartoons, sketch), open mouth, grayscale"
generator = torch.Generator(device=device).manual_seed(42)
images = pipe(
prompt=prompt,
input_id_images=input_id_images,
negative_prompt=negative_prompt,
num_images_per_prompt=1,
num_inference_steps=num_steps,
start_merge_step=10,
generator=generator,
).images[0]
gen_images.save('out_photomaker.png')
Run the following command:
python gradio_demo/app.py
You could customize this script in this file.
If you want to run it on MAC, you should follow this Instruction and then run the app.py.
- Upload more photos of the person to be customized to improve ID fidelity. If the input is Asian face(s), maybe consider adding 'Asian' before the class word, e.g.,
Asian woman img
- When stylizing, does the generated face look too realistic? Adjust the Style strength to 30-50, the larger the number, the less ID fidelity, but the stylization ability will be better. You could also try out other base models or LoRAs with good stylization effects.
- Reduce the number of generated images and sampling steps for faster speed. However, please keep in mind that reducing the sampling steps may compromise the ID fidelity.
- Demo link, run PhotoMaker on replicate.
- Demo link (style version).
- bmaltais/PhotoMaker by @bmaltais, easy to deploy PhotoMaker on Windows. The description can be found in this link.
- sdbds/PhotoMaker-for-windows by @sdbds.
- https://github.com/ZHO-ZHO-ZHO/ComfyUI-PhotoMaker
- https://github.com/StartHua/Comfyui-Mine-PhotoMaker
- https://github.com/shiimizu/ComfyUI-PhotoMaker
Provided by @Gradio
- PhotoMaker is co-hosted by Tencent ARC Lab and Nankai University MCG-NKU.
- Inspired from many excellent demos and repos, including IP-Adapter, multimodalart/Ip-Adapter-FaceID, FastComposer, and T2I-Adapter. Thanks for their great work!
- Thanks to the Venus team in Tencent PCG for their feedback and suggestions.
- Thanks to the HuggingFace team for their generous support!
This project strives to impact the domain of AI-driven image generation positively. Users are granted the freedom to create images using this tool, but they are expected to comply with local laws and utilize it responsibly. The developers do not assume any responsibility for potential misuse by users.
If you find PhotoMaker useful for your research and applications, please cite using this BibTeX:
@article{li2023photomaker,
title={PhotoMaker: Customizing Realistic Human Photos via Stacked ID Embedding},
author={Li, Zhen and Cao, Mingdeng and Wang, Xintao and Qi, Zhongang and Cheng, Ming-Ming and Shan, Ying},
booktitle={arXiv preprint arxiv:2312.04461},
year={2023}
}