AniPortrait icon indicating copy to clipboard operation
AniPortrait copied to clipboard

AniPortrait: Audio-Driven Synthesis of Photorealistic Portrait Animation

Results 109 AniPortrait issues
Sort by recently updated
recently updated
newest added

EchoMimic is capable of generating portrait videos not only by audios and facial landmarks individually, but also by a combination of both audios and selected facial landmarks. Project link: https://badtobest.github.io/echomimic.html...

感谢您们出色的工作,但我注意到一些小小的问题,我想是否能在以下几个角度和您进行讨论:1. 我注意到您们相当于使用显式关键点作为条件,在类似FOMM / LivePortrait / follow your emjio这类的工作中,都采用了一个小网络来将关键点进行隐式表示,这样是否可以降低由于drive视频检测到的关键点本身抖动带来的偏差(生成视频很抖动);2. 在Cross Attn层加入Clip特征是否有助于人脸细节的生成,或许也能降低一些推理时降噪步数;3. 先前的模型会时不时有奇怪的手生成出来,能否通过将VFHQ / Vecelvhq清洗一下进行微调来修复这个问题呢?4. 现在的模型推理时实在是太慢了,诚然,也许LCM之类的技术或许能进一步大幅降低推理时间;但出于一些别的考虑,这个模型本身是否能做的更小一些呢?

看了一下论文,我的理解是: (1)audio-driven 是用audio预测landmark序列,然后结合单张图片合成与语音相匹配的视频 (2)face reenacment是直接基于视频中的帧生成landmark序列,然后结合单帧合成与语音相匹配的视频,这里面没有音频参与,是这样吗? (3)self-driven又是什么情况? 仅仅是参考视频和要修改的图片是同一个人吗

I find in the "stage1.yaml" file `max_train_steps` is set to 300,000, and I want to ask is it really necessary to train such a long schedule? In animate anyone paper,...

https://github.com/Zejun-Yang/AniPortrait/assets/51695272/1fe11dac-3a5a-48d4-9387-49a68c92b913 作者大大,这是怎么回事? 数据问题? 训练次数不够?

python 的script执行命令行,能不能添加以下参数说明阿,生成的视频默认是左中右三段式的,怎么能生成单独的三个视频?

Hi, thanks for the open-source! I ran `/scripts/preprocess_dataset.py` but I don't think the 6d pose sequences were extracted, can you please let me know how to do this? Do you...

你好,是不是可以通过 bs 直接计算 mesh,然后找到对应的关键点,进行人脸驱动?52个表情基在哪可以找到呢?

设置H,W为256时效果特别差,需要更改相应的参数吗

It seems that Hugging Face demo is not available. I used example data. Eventually, gradio demo occurred error and result doesn't exist.