Here is a series of new face generators based on StyleGAN2, including yellow face，chinese internet-celebrity face，chinese pop-star face，world supermodel face and cute baby face generator that are improved from the old version, and two more aesthetic generators(mixed-blood face and Asian beauty face generator)are added, along with a general face attribute editor. Having made so many generators is enough, I will no longer try to make new content related to face generators, but I will explore more practical and more satisfying generation technology to better serve people(for example you can visit Video-Auto-Wipe).
The function of the generator is to provide various styles of face materials for us to use at will, and help save the cost of finding real people (faces).
The new version based on StyleGAN2 eliminates the occurrence of artifacts and distortion / damage in the picture, making the generation success rate close to 100% (see the randomly generated dataset in README.md), which can be used in mass generation tasks; in addition the quality of the pictures has been further improved, and the clarity has approached the dataset used in official training. I hope that this project will help film, television, advertising, games, and medical & aesthetic workers, and at the same time empower ordinary enthusiasts.
This project is all free and open source, I hope to help friends in need. The model is for play and research use only, and commercial use is not allowed without authorization. The model's copyright belongs to: www.seeprettyface.com. If it is helpful to you, please sponsor at the bottom ~
Do you know what the best looking human face looks like? I combined the chinese pop-star face generator with the world supermodel face generator in a carefully modulated ratio to create this mixed-blood face generator. The face synthesized by this generator has a unique and outstanding aesthetic style (the customer's evaluation is "Maquatte mask blends with the charm of Orientals"), which in my opinion is the best-looking face currently drawn by AI Generator . However, this generator has been bought out and is not open-source.
The interesting thing is that after I open-source the above generators, a chief editor of a visual magazine found me and discussed with me about whether we can make a more recognizable and "amazing" face generator-because only if AI can surpass humans in aesthetics, generation technology can effectively impact the traditional vision industry, which means that people can get the best resources at the lowest cost. What ’s more beneficial to us is that the magazine has high-quality image material resources, and I have varied training skills, so we just cooperated to make this "Asian beauty face" generator. Some of the face materials synthesized by the generator are shown below.
If you want to create a graphic, you can obtain images through this interface with a truncation rate of 0.8.
For cross-border e-commerce, a large number of customized model materials are required, such as black models:
The above-mentioned generators seem to be fully available, but in fact, this technology is still far from being truly commercialized(Model-Swap-Face might be one possibility?)... If we really want to use it to impact the traditional vision industry, there are at least two problems that need to be solved: 1. Related supporting technologies need to be improved, such as face implantation, animation generation and whole body synthesis, etc.; 2.It needs to be explored that how to build a specific generation technology service system for segmented user groups.
· Both Linux and Windows are supported. Linux is recommended for performance and compatibility reasons.
· 64-bit Python 3.6 installation. We recommend Anaconda3 with numpy 1.14.3 or newer.
· TensorFlow 1.14 or 1.15 with GPU support. The code does not support TensorFlow 2.0.
· On Windows, you need to use TensorFlow 1.14 — TensorFlow 1.15 will not work.
· One or more high-end NVIDIA GPUs, NVIDIA drivers, CUDA 10.0 toolkit and cuDNN 7.5. To reproduce the results reported in the paper, you need an NVIDIA GPU with at least 16 GB of DRAM.
· Docker users: use the provided Dockerfile to build an image with the required library dependencies.
- On Windows, the compilation requires Microsoft Visual Studio to be in PATH. We recommend installing Visual Studio Community Edition and adding into PATH using "C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Auxiliary\Build\vcvars64.bat".
My test environment configuration is: Win10, 1050Ti, CUDA 10.0, CuDNN 7.6.5, tensorflow-gpu 1.14.0, VS2017 can run perfectly.
Common problem under Windows: "Could not find MSVC/GCC/CLANG installation on this computer" How to solve it?
When installing VS2017/VS2019, be sure to select ‘Desktop development using C ++’ (shown below)
After installation, enter C:/Program Files (x86)/Microsoft Visual Studio/2019/Community/VC/Tools/MSVC/. There will be a folder with a version number. Replace the version written in dnnlib/tflib/custom_ops.py line 29(that is the version I installed) with the corresponding version number, as shown in the figure below.
1.Download the corresponding model according to the address in 'networks/download model from Google Drive.txt',and place the model in the 'networks' folder
2.Select the corresponding model and generated number in main.py and run it.
The StyleGAN2-based attribute editor (edit_photo.py) contains the same content as legacy attribute editor , containing 21 adjustable orientation for simple face attribute editing. This attribute editor applies to all generators of this project (ie yellow, internet-celebrity, star, supermodel, cute baby, mixed-blood and Asian beauty face generator) as well as official face generator.
The following examples use the yellow face generator.
Learn technical principles & Get training set: goto www.seeprettyface.com
A small sponsor if it helps you~