Video Mask by Text is retired because it could not reliably track a described subject through a clip; saved nodes remain visible as removal tombstones.
Where: Existing saved Flowgen canvases only; it is absent from Add, search, blueprints, and MCP creation.
Key ideas
- Why it was retired: Still-image segmentation was not enough to guarantee temporal identity and stable edges across an entire video.
- Saved-flow safety: A tombstone card explains the removal instead of silently deleting the node from an existing canvas.
- Choose a replacement by intent: Use Video Matte for whole foreground/chroma separation, Mask by Text for one still image, or Object Select for manual still-image selection.
Steps
- Open the saved canvas and locate the removed node.
- Record which input, output, and downstream behavior the old flow expected.
- For whole-subject video separation, add Video Matte and select Green screen or This device.
- For a single frame, add Mask by Text or Object Select.
- Reconnect and verify the replacement across the whole clip before removing the tombstone.
MCP and indexing
- create_node no longer accepts videoMaskByText.
- The retired name remains indexed so search returns this migration guidance.
- Do not suggest recreating it by manually editing saved graph JSON.
Tips
- Temporal video masking needs tracking/coherence, not merely per-frame prompting.
Limitations and important notes
- There is no automatic one-click conversion because the correct replacement depends on the original intent.
Troubleshooting
MCP rejects videoMaskByText
That is expected. Create videoMatte for video foreground extraction or maskByText for a still image.