VideoPromptMask

This is a workflow that combines Florence2 and SAM2. It automatically detects objects specified by text and outputs them directly as segmentation masks.

1. Upload the video: Load the video you want to edit into the LoadVideo node.

2. Set frame skipping for preview: Initially, set LoadVideo's select_every_nth to 10000 so you can check representative frames only, rather than processing every frame.
   Specify the mask target with text: Enter a caption for the object you want to mask in Florence2Run's text_input (e.g., "human face" for just the face, or "jacket" for just the jacket).

3. Run the workflow to check: In the rightmost node, verify that the mask matches the caption you entered.

4. Re-run with production settings: If everything looks good, set LoadVideo's select_every_nth back to 1 and re-run the workflow on all frames.

5. Export the mask video: Download the generated mask video. It will be saved in My Files under outputs.