Disable audio generation in `TI2VidTwoStagesPipeline` to improve inference speed

[AlexBlack2202]

Hi every body,

I'm currently using the TI2VidTwoStagesPipeline from the LTX-2 repository and had a question regarding audio generation during inference.

For my use case, I only need the generated video output and do not require any audio. However, it يبدو that the pipeline still initializes and processes audio-related components, which adds unnecessary overhead and slows down inference.

I’ve reviewed the implementation here:
https://github.com/Lightricks/LTX-2/blob/main/packages/ltx-pipelines/src/ltx_pipelines/ti2vid_two_stages.py

My questions:

1. Is there an official or recommended way to disable audio generation entirely in this pipeline?
2. If not, what would be the cleanest way to modify or bypass the audio stage without breaking the pipeline flow?
3. Are there any flags, configs, or modular components that can be safely removed (e.g., audio guider, audio decoder, etc.) to speed up inference?

My goal is to optimize performance (latency + GPU usage) in a production setup where only video output is required.

Any guidance or best practices would be greatly appreciated. Thanks for your work on LTX-2!

Best regards,
Alex Black