I need help with some optimal parameters for the Gradio APP I am developing

[FurkanGozukara]

APP is working atm the but I don't know if my default values are good

1. So what is the optimal image input resolution 512x512? I added face focused auto crop
3. Focuses face and expands 20%. is this good? how much percent you suggest? I am using retina face
5. Also can we change output resolution to like 768x768? if so how?

moreover it is pretty slow what optimizations can be made?

on rtx 3090 it is 6.84 second / it

perhaps update of torch diffusers etc?

one another thing is that we have a input video, re_target strategy do you suggest?

and default values for Reference Attention Weight and Audio Attention Weight

here my gradio app interface so far

also here my input video

https://github.com/tencent-ailab/V-Express/assets/19240467/c93dd2c3-8fbd-4f5c-9de5-9eec51c4dfa2

here my input face

here output video

https://github.com/tencent-ailab/V-Express/assets/19240467/aa58da83-e10f-4b34-909f-7d54246384e1

[FurkanGozukara]

@tiankuan93 @Tenvence @kuantian

[FurkanGozukara]

published my app as v1 : https://www.patreon.com/posts/105251204

1-click to auto install into a python venv

scripts for Windows, RunPod and Massed Compute - hopefully will make a free kaggle notebook too

https://github.com/tencent-ailab/V-Express/assets/19240467/635202da-a617-473f-92e9-8f1ea9fa1a8e

[cantonalex]

Can we delete this patreon crap. It goes against everything open source. The Vexpress team have open sourced the code. Don't then take it and try and monetise it by putting it on HF. Here is an open source version of HF https://huggingface.co/spaces/alexcanton/V-Express/ requires GPU for runtime