[Feature] AvatarChatbot: Allow user to choose avatar images and model type between runs, instead of just the audio question

[ctao456]

Priority

P4-Low

OS type

Ubuntu

Hardware type

Gaudi2

Running nodes

Single Node

Description

The input to the wav2lip fastAPI is currently a post request that contains only the "audio" field, the byte64str for the audio question wav2lip_server.py

Whereas the avatar image/video path and the model type are passed through environment variables, $FACE and $INFERENCE_MODE, respectively. They are set before docker run to create the wav2lip fastapi container, through entrypoint.sh. Once the container is created, they are fixed values inside the container.

We need a smarter method to merge these environment variables into the fastapi post request. We need to create a new datatype for the animation microservice in docarray.py. Then, we can use that datatype to include the fields "face" and "inference_model", and pass them to the animate function in wav2lip_server.py.

We also need to make corresponding changes in the UI, to let the user use preview images/videos, and preset model types.

The outcome of this change will be that the user can change his/her avatar to use, and the inference model type, beside just the audio question, between one run and another.