Request: Add an official Dockerfile

[AmitMY]

It would be very helpful if the repository included an official Dockerfile that captures all required dependencies for VACE and its preprocessing stack.

Right now, VACE depends on multiple external models and toolchains (e.g., Wan, GroundingDINO, SAM2, etc.), each with their own CUDA/PyTorch constraints. These versions are inconsistent across upstream projects, which makes environment setup brittle.

Some examples of version drift:
• GroundingDINO documentation mentions CUDA 11.3,
but their official Docker image runs CUDA 12.1 + PyTorch 2.1.2.
• VACE itself uses:
• Python 3.10.13
• CUDA 12.4
• PyTorch 2.5.1

A maintained Dockerfile would ensure users can reproduce the intended environment, especially when mixing:
• VACE core code
• Wan model + dependencies
• Preprocessing models (GroundingDINO, SAM2)
• FFmpeg / image libs
• Model weights + expected directory structure

Proposed solution

Add an official Dockerfile (and ideally a docker-compose.yml) that:
• Pins Python, CUDA, and PyTorch versions known to work with VACE.
• Installs GroundingDINO, SAM2, and other preprocessing packages.
• Documents where users should mount or place model weights.
• Optionally includes a simple entrypoint script.

This would massively reduce environment inconsistencies and speed up onboarding.

Happy to help with an initial draft

[AmitMY]

Here's my drafts: https://github.com/sign-language-processing/reproduction/tree/main/repositories/ali-vilab/VACE