Tencent AI Lab - V-Express Image to Animation Gradio Web APP and 1-Click Installers for Windows, Massed Compute, RunPod and Kaggle (Patreon)

Patreon exclusive posts index

Join discord to get help, chat, discuss and also tell me your discord username to get your special rank : SECourses Discord

V-Express: Conditional Dropout for Progressive Training of Portrait Video Generation : https://github.com/tencent-ailab/V-Express

21 June 2024 Update

• Scripts upgraded to Version 7

• Download latest V_Express_Installers zip file

• Huge VRAM optimizations have arrived

• Now with 8GB VRAM GPUs you can even generate 20 second videos or maybe even more

• Now on Kaggle you may even generate 1 minute video try it

• Installers are updated and all model files will be downloaded as a single zip file

• Thus, you will be able to resume model download if fails with Resume_Models_Download_If_Error_Occurs.bat file

• So pay attention to the installer CMD

13 June 2024 Update

• All installers updated since the original authors had changed model source folder structure

12 June 2024 Update

• Full Cloud tutorial video published : https://www.youtube.com/watch?v=GXBiqJOc9FE

• Check the video chapters for Massed Compute, RunPod and Kaggle

• Please also consider upvote and a comment to below reddit thread:

• https://www.reddit.com/r/StableDiffusion/comments/1ddsm72/tutorial_for_how_to_install_and_use_vexpress/

7 June 2024 Update

• Free Kaggle Account notebook added with instructions

• The notebook file is inside attachments zip file

6 June 2024 Update

• Full Windows tutorial video published : https://youtu.be/xLqDTVWUSec

• Please also consider upvote and a comment to below reddit thread:

• https://www.reddit.com/r/StableDiffusion/comments/1d94wv8/vexpress_1click_ai_avatar_talking_heads_video/

Works on Windows, Massed Compute and RunPod. Hopefully will try to make a Kaggle notebook as well.

Follow instructions for RunPod and Massed Compute. For RunPod Cuda 11.8 having template mandatory for high speed

You can see 34 comparison tests results and their configs in this file : https://huggingface.co/MonsterMMORPG/SECourses/resolve/main/Comparison_Tests.zip

I think best config is as below but you should look at each test

• Retarget Strategy = offset_retarget , Reference Attention Weight = 1 , Audio Attention Weight = 1 and rest is default

• Lower Audio Attention Weight = lesser erroneous mouth movements

Requirements:

• You need Python 3.10.11, Git, FFmpeg, CUDA 11.8, and C++ tools

• How to install all above is shown in below tutorial step by step

• https://youtu.be/-NjNy7afOQ0

• If you still can't make it upgrade membership and message me from Discord

How To Install

• I am recording a tutorial right now but so straight forward

• Just double click and run Windows_Install .bat file

How To Use

The generated files will be saved inside outputs folder inside V-Express

As you increase your input audio or video duration, it will become slower for each step

I don't know if it has a duration limit but i generated up to 20 seconds very easily

You can use Massed Compute A6000 - 48 GB GPU if you get out of VRAM error - 31 cents per hour