aihacker111 / Efficient-Live-Portrait

Fast running Live Portrait with TensorRT and ONNX models

MIT License

119 stars 9 forks source link

Open aihacker111 opened 1 month ago

aihacker111 commented 1 month ago

Check List:

This new update is using Face-ID AcrFace to control Face Animations in the Multiples Images
I'm also update new TensorRT run on latest TensorRT version
Project can run on Google Colab for demo, I'll update Gradio Interface later
In many near day ago, I'm assigned to the new research paper, so I cannot update new feature faster, but now the model is came to the training process, so I come back to update new idea on this great project Thank you and see result using Face-ID Adapter in Live-Portrait
Checkout here for run demo: https://github.com/aihacker111/Efficient-Live-Portrait

368220873_826368889022136_4472311944594836999_n

x4080 commented 1 month ago

Hi, I cannot seems to play the video ? Is it only me ?

BTW did you implement the video 2 video yet ?

aihacker111 commented 1 month ago

@x4080 seem like only you cannot open video , v2v , I’m plan for this

aihacker111 commented 1 month ago

@x4080 BTW , I’m also work on multiple object in driving video , because the official is allow only 1 face video , but we can do more than 1 face

municef1 commented 1 month ago

Thanks to your great work!

How much difference is there in inference time compared to the existing model pipeline in practical terms?

aihacker111 commented 1 month ago

@municef1 TensorRT FP16 is work with 374 frame is 30 second and fp32 is 87s , result of fp32 is better than fp16

x4080 commented 1 month ago

@x4080 BTW , I’m also work on multiple object in driving video , because the official is allow only 1 face video , but we can do more than 1 face

Cool ! Cant wait for it

aihacker111 commented 1 month ago

@x4080 support me in the adding audio bro, I'm take time to fixing the output, seem like it cannot return out of the loop

aihacker111 commented 1 month ago

@x4080 Check out this my friend

aihacker111 commented 1 month ago

@x4080 Video2Video is now availabled, you can use it, very funny

x4080 commented 1 month ago

@aihacker111 Thanks, very cool man

aihacker111 commented 1 month ago

@x4080 Hey man, I'm update controlnet open-pose feat sdxl lightning text+image to image feature, you can try on this bro

x4080 commented 1 month ago

@aihacker111 yes thats very good, is it photo to sdxl using openpose then live portrait ?