Nodes

Every node on a canvas is one of the types below. A node is either a source (it holds something you gave it), a generator (it produces something when you press Play), or both, depending on what you wire into it.

Where a node has a keyboard shortcut for adding it, that key is noted at the end of its section.

Text

A note you write, or the place an Assistant writes its result. A text node has no model of its own — it is just text, edited in place with a rich editor.

• Feeds: the prompt input of any generator, and the text context of an Assistant.
• Takes: nothing, unless an Assistant is wired into it, which turns it into that Assistant's output.

Text is what supplies prompts. An image generator with no text on its prompt input has nothing to generate from.

Shortcut: 1

Assistant

Transforms what you connect to it. The Assistant reads its connected input — text, images, video, audio, or a list — and writes a result into a Text or List node downstream.

• Modes: Enhance prompt, plus your own instruction in the instructions field.
• Output: choose Text to create or update a connected text node, or List to produce several items at once.
• Takes: text on the prompt input, and media on the image, video and audio inputs.

The Assistant has no result of its own and no prompt editor. Connect an input, choose where the output should land, then run it.

Image generator

Generates an image from a prompt, optionally using other images as reference.

• Takes: a text or Assistant node on the prompt input; images on the reference input, including a frame pulled out of a video.
• Settings: aspect ratio, resolution and quality, where the model supports them. The available values come from the selected model, so some models offer none.
• Produces: an image, editable in place with crop, relight, restyle, recolour and reframe.

An image node that holds an upload or a Stocks asset instead is a source: wire it into another image node's reference input to generate from it.

Shortcut: 2

Video generator

Generates video. Which media inputs a video node offers depends on the model, so pick the model first — the handles appear once the model that supports them is selected.

• Takes: a prompt, plus any of a start frame, an end frame, reference images, a reference video and a reference audio track, subject to the model.
• Settings: aspect ratio, resolution, duration and whether audio is generated, read live from the model's capabilities.
• Produces: a video, plus its first and last frame as separate image outputs you can wire onward.

Videos can be trimmed and upscaled on the canvas.

Shortcut: 3

Audio generator

Generates speech, sound effects or music from a prompt.

• Takes: a prompt, and nothing else.
• Settings: voice for speech models, duration and loop for sound effects, length for music. Each model lists its own allowed values.
• Produces: audio, which can be wired into a video node as its track where the video model accepts one.

Shortcut: 4

List

Holds many items of one kind — text, images or videos — and is how you do the same thing many times.

Wire a list into the prompt input of an image or video generator and that generator fans out: one run per item, in parallel, with one result per item written into a list on its output.

• Takes: one producer — the Assistant, image or video node that fills it, matching its kind.
• Has no model or settings. The node it feeds owns those.
• Re-running is safe: only the items that run produced are replaced. Your uploads and per-item edits stay.

Text and image lists can drive an image or video generator. A video list can only drive a video node.

Shortcut: 5

Storyboard

Generates a grid of related panels from one prompt, so a sequence stays consistent instead of drifting between separate runs.

• Grids: 2×2, 3×3 or 4×4.
• Takes: a prompt, and reference images to keep the panels on-model.
• Produces: the grid, plus one image node per panel laid out beside it, so each panel can be edited or wired onward on its own.

Shortcut: 6

Reframe

Re-renders an image you already have from a different camera position, keeping the subject, lighting and style identical and changing only the framing.

• Angles: twelve presets, including 3/4 view, aerial, back view, closeup, extreme closeup and extreme long shot. Pick several and each becomes its own result.
• Aspect ratio: match the input, or 16:9, 9:16, 1:1, 4:3 or 3:4.
• Takes: an image.

Shortcut: 7

Infographic

Turns a product image into laid-out marketing artwork — headline, supporting copy and typographic treatment composed onto the image.

• Styles: ten presets, from editorial and minimal caption through feature callout, spec panel and graphic poster. Each selected style becomes its own result.
• Engines: Integrated renders the type with the image; the two Crisp engines render type separately for sharper glyphs.
• Placement: let it use existing empty space, or expand the canvas left, right, up or down and set the type in the new area.
• Resolution: 1K, 2K or 4K.
• Takes: an image, and optionally a text node on the copy input carrying your headline and supporting line. With no copy connected it writes its own from what the image visibly shows.

Shortcut: 8

How edges work

An edge with no named target is the prompt edge, and it carries text. Named inputs carry media instead: reference images into an image node, start frame, end frame, references, video and audio into a video node, and image, video and audio context into an Assistant.

When you drag a connection, the handle is chosen for you from the two node types — an image dropped on a video node becomes its start frame, an audio node dropped on a video node becomes its audio track. Connections the target cannot accept are refused rather than silently ignored.

A video node has three outputs rather than one: the video itself, its start frame and its end frame. The frames behave as images everywhere an image is accepted.

Nodes you may still see

A camera angle node is no longer in the toolbar — reframe covers the same ground with more angles. Canvases that already contain one keep working.