AI face-changing software offline version, the software is developed based on the open-source projects rope and facefusion, supporting Mac and Windows. Interested students can download and learn, please do not misuse this technology and comply with laws and regulations!
-
V3.0 Download
Win10, Win11 V3.0
(1) Quark Cloud Disk link: https://pan.quark.cn/s/cadcdc64e501
Extraction code: ZUpY
(2) Baidu Cloud Disk link: https://pan.baidu.com/s/1uhpL3x-FWAqQBJ86C6WBvg
Extraction code: u6rn
Mac Apple computer version V1.4
(1) Baidu Cloud Disk link: https://pan.baidu.com/s/122lA1xV4oipOBghGE8feYg
Extraction code: k5g6
(2) Tianyi Cloud Disk link: https://cloud.189.cn/t/by2yqafuIjiu (Access code: bn1n) -
Contact information
Bilibili video: https://www.bilibili.com/video/BV1pu411T7R4 -
Update records
The software is based on the open-source projects rope and facefusion.
V3.0 version: 2023-10-14
- Facefusion updated to 1.3.1, roop synchronized update;
- The face repair model now has gfpgan1.2-1.4, codeformer, gpen models, all using onnx models. Compared with version 2.3, gfpgan has a speed increase of 3-5 times, and the face repair speed is the fastest, followed by gpen, and codeformer is the slowest;
- The face-changing model supports the selection of fp16, but many feedbacks have black image phenomenon with fp16, but it is faster. By default, it does not come with fp16;
- The background HD repair also supports x2, x4 models. The x2 model has a smaller memory usage, but the effect is not as good as x4;
- The HD repair mixing degree parameter ranges from 0 to 100, with a default of 100, which represents the highest repair program;
- With the 3-5 times improvement in face repair, the live face-changing has reduced stuttering with face repair enabled;
- Fixed a bug that caused a large frame rate for video parsing;
- Now the intermediate frame cache directory is located in the software directory/cache_file folder, and there is a button on the main interface to open the cache directory.