Graph

Workspace model rules filter model and operation choices. An existing node that uses a blocked model shows a workspace-admin refusal when run. Ask a workspace admin to review Settings > Security & compliance; previous outputs remain available.

The Graph is where you build a shot out of connected steps. Each box is a node that does one job, and the wires between them carry pictures, clips and sound from one to the next.

The reason to use it over the Create panel is reuse. Change the picture at the start of a chain and everything downstream can be re-run from it, instead of you repeating five steps by hand.

The page is headed Node Graph. Each shot has its own graphs. Pick the shot and the graph in the strip at the top. New shot and New graph each ask for a name, which you can type or leave blank, then make it when you press Create.

1. Add nodes from the palette on the left, or drag a file onto the canvas to start from it. On an empty canvas, Start from template adds a ready-wired chain.
2. Drag from the dot on one node to an input on another to wire them together.
3. Click a node to open the Inspector and set it up.
4. Press the run button on a node to make it work.
5. Finish a chain with an Output node to mark the result as the keeper.

A node keeps every result it makes. Press All results under the node, then click a picture or video to view or watch it. From there, Approve marks that take as approved, and Use this makes it the result the node passes on.

The small map in the corner shows the whole graph at once. Colours tell you what is happening: amber for running or waiting, green for done, red for failed, grey for not run yet.

The palette groups the nodes into families: Source, Create, Alter, Sound, Free edits, Text and Finish. Free edits change pictures in your browser and cost no credits. Keyboard shortcuts (?), at the top, lists the keys; pressing ? opens it too.

• Build & Run, at the foot of the palette, runs the whole graph. When two or more nodes are selected, Run selected runs only those, with their price under it.
• Templates: select nodes and press Save selection to keep them as a template, for Only me or Everyone in org. Click a saved one to insert it.
• Community templates ▾ inserts one of the ready-made graphs from Community.
• Add whole graph to Community sends the whole graph, not just your selection. Name it and press Submit for review; it shows in the gallery once we approve it.
• Publish as app turns the graph into a form on the project's Apps page.

On a Video Gen node set to Seedance 2.5 at 1080p, tick Try a cheap 480p first to make a low-cost test before the full render. If you like the test, press Make it 1080p on the node. Until you do, nodes wired after it wait for the 1080p.

On a Video Gen node set to Kling 2.6, a last frame makes a clip without sound, with or without a first frame. While a last frame is wired, Sound shows Off and can't be changed. Unwire the last frame to turn sound on again.

Kling v3, Kling 2.6 and Kling 2.5 Turbo need a prompt even with a first or last frame wired in. Under What should happen in this clip?, describe the motion or camera movement, for example: The person turns toward the camera and smiles. The Inspector's run button stays unavailable until the prompt has text, and a blank prompt blocks every run before credits are reserved. A wired text prompt can supply the description too.

Character Replacement needs a driving video up to 30 seconds and a character image. If the video is too long, use Trim before running it. Trim offers 29.8 seconds to leave room for encoding rounding; even a clip measured at 30.03 seconds needs trimming. Use footage with a person's face visible throughout. If a face cannot be detected in part of the clip, the result asks you to choose suitable footage and releases the reserved credits. Output covers up to about 16 seconds at 24fps or 13 seconds at 30fps; the rest is trimmed.

Kling models need pictures at least 300 px on each side. A Kling first frame, and any picture wired into Kling Omni, can't be more than 2.5 times as wide as it is tall, or as tall as it is wide. The node turns such a picture down before any credits are used, and says which file to replace.

The expand button on a prompt box opens a larger editor. It lists the node's connected inputs beside the text, with their pictures, so you can see which is which while you write. Type @ in any prompt box to pick one. It puts in the name the model reads, such as @Image1 or Image 2.

A Reference node's picker shows each file's name and its real size, such as 4096×2160, so you can tell HD and 4K versions of a shot apart. An Output node can export an uploaded clip or picture straight from a Reference node, as well as a take.

Project assets, under Reference in the palette, lists every picture, video and audio file in the project, from every shot. Search by name, or narrow the list by type, by shot, to General or Unfiled files, or to files or takes. Drag one onto the canvas to make a Reference node with it, or onto a Reference node to swap its file. Add does the same without dragging, and Use in selected puts the file in the Reference node you have selected. When previews are on for your workspace, a 3D type lists the project's 3D files with their pictures. Click one to watch it turn. A 3D file can't be dragged onto the canvas.

If a node's picture or clip can't load, the node says why. You're signed out or Couldn't load media means your sign-in or connection dropped; the node loads again by itself once you're back, or reload the page. This file isn't available means it was removed, is still processing, or is in a project you can't open, so re-attach it or pick another file.

Your graph saves as you work. You can close the tab and come back to it. If a save fails, the canvas says so; press Try again to save it again.