Give the chat assistant inputs two ways: attach images/videos with a role (reference, first frame, last frame) or select existing canvas nodes — both get auto-wired into the node(s) the assistant creates.
Where: Assistant panel → paperclip / canvas selection — Attach media with a role, or select source nodes before sending
Key ideas
- Attachment roles: Each attached image/video carries a role you set in the chip dropdown: Reference, First frame, or Last frame. Reference images feed every image you generate; first/last frames feed the video producer.
- Role-aware handles: Roles map to real input handles on the target: first→start_img, last→end_img, reference→ref_img on video/avatar nodes (these appear once a Veo model that supports them is selected), or the node's content/reference image input for image targets.
- Canvas selection as input: Nodes you select on the canvas are detected by the panel and wired into the first node the assistant creates; selected images spread across a video target's frame slots (start → end → reference) instead of all landing on one.
Steps
- In the Assistant, click the paperclip and choose one or more images/videos (they appear as chips).
- Set each chip's role in its dropdown — Reference, First frame, or Last frame.
- Optionally select existing nodes on the canvas; the panel shows 'N canvas nodes selected — will be wired as input'.
- Type your description and send; the assistant creates input nodes for your attachments, labels them by role, and wires both attachments and selected nodes into the new node(s).
- After it runs, attachments are consumed (cleared) since they're now wired into the graph.
FAQ
Do I need to draw the edges for my attachments myself?
No. Attachments and canvas-selected nodes are auto-wired. Reference images go to every image you create; first/last-frame images go to the video producer; selected nodes feed the first created node.
Why did attaching a last-frame image switch the video to a bigger model?
The frame-role handles (start/end/reference) only exist on Veo models that support them, so the assistant selects veo-3.1 (not lite) when a last frame or reference is present so those slots appear.