Feature Request: FlashVSR Support

[edisonchan]

Is your feature request related to a problem? Please describe.

Currently, Video2X does not support FlashVSR, a modern video super-resolution framework. This limits users who want to leverage FlashVSR’s speed and quality improvements for long video upscaling. I’m often frustrated when I need to manually run FlashVSR outside of Video2X, which breaks the streamlined workflow.

Describe the solution you'd like
I would like Video2X to integrate FlashVSR as an optional backend for video upscaling. Ideally:

• Allow users to select FlashVSR in the GUI/CLI as an upscale engine.
• Support chunked video processing to handle long videos efficiently.
• Ensure compatibility with existing Video2X pipeline (input/output formats, logging, progress reporting).

Describe alternatives you've considered

• Running FlashVSR separately and then combining results manually. This is cumbersome and error-prone.
• Using other engines like Waifu2x or Real-CUGAN, but they don’t match FlashVSR’s performance for certain video types.

Additional context
FlashVSR repo: OpenImagingLab/FlashVSR
Example: https://github.com/zji996/FlashVSR-app
FlashVSR provides GPU-accelerated video super-resolution with chunked loading, making it suitable for integration into Video2X.
A modular design (e.g., FlashVSREngine) could keep the codebase clean and maintainable.

[k4yt3x]

The primary blocking factor is that v2x is now using ncnn-Vulkan backends written by nihui, like: https://github.com/nihui/realcugan-ncnn-vulkan. If nihui doesn't write an implementation for FlashVSR it's difficult to integrate this new model. This could be easier if we rewrite using say libtorch so we can use .pth models.