Comfyui Generate Digital Talking Human Workflow

################## AI大师工作流 ##################

MuseTalk，是腾讯开源的说话数字人项目。目前，经过各开源作者们的努力改造移植，已经可以在ComfyUI中直接用图形节点化方式操作，生成会说话的数字人。其特点是：

①、默认训练模型就已经是256*256的脸部像素尺寸，要比Wav2lip方案的模型高清很多。

②、同我之前用过的ER-NeRF, Geneface++，RAD-NeRF等NeRF类的数字人项目对比，这些项目都需要懂一些代码才能部署，而ComfyUI-MuseTalk仅需在首次安装时稍微接触点代码（可无脑复制），即可成功运行。

③、同时，与上述NeRF类的数字人对比，该模型又具备很不错的泛化性，无需再单独为每个人定制训练模型。

④、如果再与我用过的SadTalker, Video-Retalking等数字人项目对比，ComfyUI-MuseTalk不仅支持输出高清模型，其运行速度也比这些快很多。

⑤、生成高清数字人的工作流【完整版详细教程】网址：

https://di2oc32s4h8.feishu.cn/wiki/WnawwDvXniUj6UkILuHcr7g5nYd?from=from_copylink

################## AI大师工作流 ##################

1、（可选项目）先使用我之前的分享的MuseV工作流将图片生成视频，或者直接选择一段高清面对镜头的说话人视频。

2、将视频帧率处理为25fps，便于适配原生训练模型的帧率。

3、帧数读取上限可以按你的需求自己调整（如果原始视频越长，帧数读取值越多，对电脑硬件要求越高）

4、MuseTalk模型下载地址：百度网盘：链接: https://pan.baidu.com/s/1ME-3fhP1IwSoq0DBumMvsA?pwd=x9a2 提取码: x9a2

5、欢迎关注：AI大师工作流。抖音/B站/小红书/视频号/快手等平台均同名。

https://www.douyin.com/user/MS4wLjABAAAAviq9ixG7tShWv_AJNEvCqlwZXd8YRTyCygSNpZ7J0aQ

################## AI大师工作流 ##################

################## AI_master_workflow ############

MuseTalk, is Tencent open source to Digital Talking Human project. At present, through the efforts of various open source authors to transform and transplant, it has been possible to directly use in a graphical node-based manner in ComfyUI to generate  digital talking human. Its characteristics are:

① The default training model is already 256*256 face pixel size, which is much higher than the model of the Wav2lip scheme.

②. Compared with the NeRF digital human projects such as ER-NeRF, Geneface ++,RAD-NeRF, etc. that I have used before, these projects all need to know some code to be deployed, while ComfyUI-MuseTalk only need to touch the code a little (which can be copied without brains) during the first installation to run successfully.

③, At the same time, compared with the above-mentioned NeRF class of digital humans, the model has a very good generalization, no longer need to customize the training model for each individual.

④ If compared with the SadTalker, Video-Retalking and other digital projects I have used, ComfyUI-MuseTalk not only supports the output of high-definition models, but also runs much faster than these.

⑤ Workflow for Generating High Definition Digital Human [Full Detailed Tutorial] Website:

https://di2oc32s4h8.feishu.cn/wiki/WnawwDvXniUj6UkILuHcr7g5nYd?from=from_copylink

(Only the Chinese version, but you can use the browser's translation function to read. )

################## AI_master_workflow ############

1. （Options） Use the MuseV workflow I shared earlier to generate a video from the image, or directly choose a high-definition speaker video facing the camera.

2, the video frame rate is processed as 25fps, which is easy to adapt the frame rate of the native training model.

3, the frame reading limit can be adjusted according to your needs (if the original video is longer, the more the frame reading value, the higher the requirements for computer hardware)

4，MuseTalk model download site :https://github.com/TMElyralab/MuseTalk?tab=readme-ov-file#download-weights.

5, Could you GIVE me a star or follow me on:Douyin /Bilibili /LittleRedBook/WeChat video/KuaiShou?

All these platforms are the same name.

https://www.douyin.com/user/MS4wLjABAAAAviq9ixG7tShWv_AJNEvCqlwZXd8YRTyCygSNpZ7J0aQ

################## AI_master_workflow ############

参考节点/Reference Node：

https://github.com/chaojie/ComfyUI-MuseTalk

https://github.com/TMElyralab/MuseTalk

Thanks to the open source community, the world is better because of open source