How nodes connect

Everything in Node Composer happens by wiring one node’s output into another node’s input. Once you can read the handles, the whole app makes sense — so start here.

Handles are rings

Inputs sit on the left edge of a node, outputs on the right. Each handle is a small ring, and it tells you two things: its colour is the kind of data it carries, and the letter or number inside it is which slot it is.

Green carries a text prompt, blue carries an image, amber carries a video, and there’s a separate blue handle for audio. A ring is hollow while nothing is connected, and fills with colour once it is — the quickest way to see at a glance what a node still needs.

The letters follow a simple pattern: P for prompt, I for image, V for video, A for audio, and F / L for first frame and last frame on video models that support it. Numbered slots (1, 2, 3…) are for reference images, and the order matters — it’s the order the images are sent to the model. Hover any ring to see what it is, and hover a connected image or video slot to see a thumbnail of what’s wired in.

Slots appear as you fill them

Nodes that take several reference images don’t show ten empty rings up front. Instead you see the slots you’ve filled, plus one empty ring waiting — an I on image inputs, hollow until something lands in it. Connect something and it becomes the next numbered slot, with a fresh empty I appearing below. Delete a slot in the middle and the rest renumber automatically, so there’s never a gap. You can also drag a ring to reorder it, or pull it sideways to disconnect it.

What can plug into what

Prompt handles accept a Prompt Generator, a Text Area, a Prompt Split, or a finished result’s own prompt handle. Image and video handles accept the matching media type, a Media Group, or a Media Placeholder. Audio handles only accept an audio file.

If you try to connect something invalid, the wire simply won’t land — no error, nothing breaks. And if a wire does connect but the current model can’t use it — usually because you’ve added more reference images than the model accepts — the connection stays in place but goes dashed and faded, so switching back to a model that supports it brings it right back to life.

The result handle

Every finished image or video carries its own small output handle. Wire it into another generator’s prompt input, and you get the original prompt back — plus the exact model and settings that made it. It’s the fastest way to riff on something you’ve already made.

How a prompt flows

Wire a Prompt Generator into an Image or Video Generator’s prompt input, and the prompt box becomes read-only, showing the upstream text live. Re-run the Prompt Generator and the downstream text updates immediately — it’s a live connection, not a copy. Disconnect it, and the box goes back to whatever you’d typed there before.

If the upstream node is still working when you hit Generate, the app won’t fire with an empty prompt — it waits and starts automatically once the prompt is ready.

Now that you can read a graph, see Combining nodes into a workflow for recipes you can build.