Normalize multimodal input keys in AnyToAnyPipeline (#47074)
* Keep AnyToAnyPipeline video inputs plural
Signed-off-by: Ting Sun <suntcrick@gmail.com>
* Run AnyToAnyPipeline video test with real model generate
Signed-off-by: Ting Sun <suntcrick@gmail.com>
* style
---------
Signed-off-by: Ting Sun <suntcrick@gmail.com>
Co-authored-by: Raushan Turganbay <raushan@huggingface.co>