Workflow nodes

Every node you can add to a workflow, grouped the way the add-node picker groups them, with what each one takes in and what it gives back. Use it when you are building or changing a workflow and need to know which node does the job.

How nodes fit together

A workflow reads left to right. Inputs hold the prompts, images and other material you start from. Generators make new images, video, 3D meshes, audio and text. Editors change something that already exists, utilities reshape, combine and route files without AI, and judges filter a set down to the items that pass your bar.

You connect a node's output handle to another node's input handle. Each handle is coloured by the kind of data it carries (text, image, video, audio, mesh, reference set), and only matching handles connect. Drag from an output and drop it on empty canvas to open the picker filtered to the nodes that accept it.

Nodes work on lists. When a node receives several values on one input, it runs once per value, and when it receives lists on several inputs, it runs once per combination. Three prompts into a generator that makes four images each gives twelve images. This is how one workflow produces a whole set in a single run, and it is also how a run gets more expensive than you expected, so check the cost on Run before you start.

To add a node, use the buttons along the bottom of the canvas (Generators, Editors, Judges, Utilities, Inputs, or a media type), right-click empty canvas, or press ⌘K and choose Add node. The search box finds any node by name, whichever group you opened.

Inputs

Input nodes hold what you type or upload and pass it on unchanged. They are the only nodes that can become workflow inputs, the fields a person fills in when they run the workflow. See Find and run a workflow for how inputs look to the person running it.

Images

A list of images. Connect it to a generator to guide composition, pose or style, or to an editor as the images to change.

Videos

A list of videos. Connect it to a video generator, a video editor or a video utility.

Reference Sets

A list of reference sets, each with an optional weight. Connect it to a generator to apply their style, character or other learned look. See Reference sets.

Text

A list of plain text values: prompts, names, captions or any other words a later node needs. Each value you add is one item, so a Text node with five prompts runs the connected generator five times.

Meshes

A list of 3D mesh files. Connect it to the 3D editors to animate, retexture, rig or remesh them.

Audio

A list of audio clips. Connect it to nodes that take audio, such as Mix Video, Lip Sync Video or the audio editors.

Numbers

A list of whole numbers, for counts, seeds or dimensions a later node reads.

Decimals

A list of decimal values, for weights, strengths and scales such as the weights in Adjust Similarities.

Vector Images

A list of SVG files. Pass them through Rasterize before a node that needs pixels.

Generators

Generators make new media. Each one has a model picker; the model you choose decides which sizes, durations and options the node offers. See Choose a model.

Generate Image

Makes images from a prompt, reference images, reference sets or any mix of them. Set the model, the image size and how many images each run makes. Optional image inputs steer the result: an outline, structure, line art or pose image, a source image, or a style reference. Type @ in the prompt to mention a reference set by name; it is added to the node's reference sets. Every image the node makes has the size you set, so to get several aspect ratios from one image, follow it with Reframe Image, Resize Image or Layered Composition.

Generate Video

Makes videos from a prompt, images or both. Connect an image to First Frame to start the clip from it, and on models that support it, to Last Frame to end on it. Duration, size and whether the model also generates audio depend on the model. Match the first frame's aspect ratio to the video size, or the frame is cropped or stretched to fit.

Generate Text

Writes freeform text from your instructions with a language model, for example product descriptions, dialogue lines or alt text. Models that read images, video or audio can take them as context.

Generate Prompts

Writes new prompts from the prompts you give it, for variations on one idea. Be specific: "cute cat sitting down, flat icon style" gives more useful variations than "cat". The number of prompts comes from the node's batch size, not from a number in your instruction. Each request is answered on its own; the node does not remember earlier runs or hold a conversation.

Generate 3D Mesh

Makes a 3D mesh from a prompt or a source image. Connect it to the 3D editors to rig, animate, retexture or remesh the result.

Generate Audio

Makes speech, sound effects or music from a prompt. It can also remix existing audio or score a soundtrack for a video, depending on the model.

Image editors

Edit Image

Changes existing images to match a prompt: clean-ups, variations and targeted changes that keep the original composition. Connect the images to change to Image to Edit. Add images that show what you want under Editing References, in named groups. A group set to Use as references sends all its images to the model together; a group set to Use as variants runs the edit once per image in it, so four images to edit with a two-image variant group gives eight results. Size set to Auto keeps each source image's size.

Upscale Image

Raises an image's resolution while keeping its detail, for exports, storefronts and large formats. Some models offer creativity and resemblance controls for how much new detail they add. To make an image smaller, use Resize Image.

Remove Background

Cuts the subject out onto transparency at the image's original size. Connect an image and it is ready to run.

Reframe Image

Changes an image's aspect ratio, filling the new area with generated content. A prompt can describe what the new area should show.

Split Into Layers

Breaks a finished image into separately editable layers and delivers them as one layered PSD. Follow it with Explode Layers to get each layer as its own image.

Vectorize

Converts images to SVG, for icons, logos and other assets that must scale without losing quality.

Segment Image

Cuts out the part of an image you describe in words, onto transparency, for compositing.

Video editors

Edit Video

Changes an existing video to match a prompt, with optional reference images in named groups. The model limits how long the input video can be; split longer footage with Trim Video first.

Reframe Video

Changes a video's aspect ratio, filling the new area with generated content. Generating at the aspect ratio you need gives better results than reframing afterwards, so reframe only footage you cannot regenerate.

Remove Video Background

Cuts the subject out of a video, keeping its audio, frame rate and duration.

Upscale Video

Raises a video's resolution, keeping its audio, frame rate and duration.

Interpolate

Generates in-between frames to raise a video's frame rate or stretch it into slow motion.

Lip Sync Video

Re-syncs the mouth in existing footage to a new audio track, for dubbing and localisation.

Lip Sync Image

Animates a still portrait into a talking clip driven by an audio track. The clip is as long as the audio.

3D editors

Rig Mesh

Adds a skeleton to a mesh so it can be posed and animated.

Animate Mesh

Animates a rigged mesh. Choose the animation with a connected reference set.

Retexture Mesh

Generates a new surface texture from a description, leaving the geometry as it is.

Remesh

Rebuilds a mesh's topology: converts it to quads, caps its face count, or both, for cleaner and lighter geometry.

Segment Mesh

Splits a mesh into separate parts, by body-part templates for characters or by granularity for any model.

Audio editors

Change Voice

Replaces the voice in a recording with a voice from a reference set, keeping the original delivery and timing.

Isolate Audio

Keeps the main voice or sound in a recording and drops the background and accompaniment.

Remove Noise

Removes background noise, hiss and hum from a recording.

Separate Audio

Splits a recording into two tracks: the sound your prompt describes, and everything else.

Judges

A judge scores every item it receives against a rubric you write, using a language model, and passes on only the items that clear the bar, unchanged. Put one after a generator to stop weak results reaching the rest of the workflow. Write the rubric as the checks a reviewer would make: "the character faces the camera", "no text in the image".

Smart Image Judge

Scores images and layered PSDs.

Smart Video Judge

Scores videos.

Smart Mesh Judge

Scores 3D meshes.

Smart Text Judge

Scores text, such as generated prompts before they reach a generator.

Text utilities

Prompt Template

Turns a prompt into a template. Type @ and a name to make a placeholder, for example "@character holding a @weapon"; each placeholder becomes an input you connect a list of values to. The node writes one prompt for every combination of values, so three characters and four weapons give twelve prompts. Several templates in one node can share the same placeholders.

Concatenate Text

Joins texts from several inputs, in order, with a separator between them (a space unless you set another, such as ", " or " and then "). Every combination is produced: "Fish" and "Sharks" joined with "are cute" and "are scary" give four sentences.

Substitute Text

Fills the placeholders in one template with the texts connected to it, one input per placeholder.

Image utilities

Layered Composition

A layered layout for ad creatives, store visuals and other assets built from several images. Choose the output sizes, add layers, and place each layer separately for each size; then connect generators to fill the layers, so one layout makes many variations. Select Edit composition on the node to open the layout. It outputs both a flattened image and a layered PSD. Typical uses: putting a square asset onto a landscape background, placing a logo over key art, and previewing an app icon with rounded corners.

Resize Image

Resizes images to a target resolution, with options to fit, crop or pad.

Render Text

Draws text onto images, with a choice of font, size, colour and alignment.

Explode Layers

Turns a layered file into one image per visible layer, topmost first, so each layer can go down its own branch.

Slice Image

Cuts a sprite sheet or multi-view sheet into tiles, in reading order.

Select Images

Picks one image, or a range, out of a list by position. Negative positions count from the end, so -1 is the last image.

Image Grid

Lays several images out on one contact sheet, so a model can see several poses or views in a single image.

Adjust Similarities

Sets how strongly each reference image influences a generation.

Rasterize

Renders SVG files to images at a fixed size, for nodes that need pixels.

Extract Frame

Takes a still from a video: the first, middle or last frame, or the frame at a time you set. Feed the last frame of one clip into the First Frame of the next Generate Video to chain clips into a continuous sequence.

Smart Extract Frame

Finds the frames that match a moment you describe, such as "the moment the sword hits", and extracts them from each video.

Video utilities

Video Editor

Arranges video, image and audio clips on a timeline and renders them into one video. Select Open video editor on the node to lay out the timeline.

Stitch Videos

Plays two or more videos back to back as one video.

Trim Video

Keeps the part of each video between a start and an end time, in seconds. Every connected video is trimmed to the same times.

Image Sequence to Video

Plays an ordered list of images as a clip, with a frame rate, looping and ping-pong, for example to animate sliced sprite frames.

Mix Video

Puts audio tracks onto videos, one result for each video and audio pair.

Add Subtitles

Transcribes a video and burns in animated subtitles.

Extract Audio

Takes the audio track out of a video.

Split Video & Audio

Separates videos into a video-only and an audio-only file each.

Resize Video

Resizes videos to a target resolution, with options to fit, crop or pad.

Adjust Video Speed

Speeds up or slows down videos without changing the pitch of their audio, at a constant speed or with a smooth ramp.

Adjust Hue & Saturation

Shifts the colours and their intensity in videos.

Audio utilities

Trim Audio

Keeps the part of each audio file between a start and an end time, in seconds.

Adjust Audio Volume

Raises or lowers the volume in decibels, with optional fade-in and fade-out, and can normalise loudness to the EBU R128 broadcast standard so a set of clips plays at the same level.

Adjust Audio Speed

Speeds up or slows down audio without changing its pitch, at a constant speed or with a smooth ramp.

Mix Audio

Mixes several audio tracks into one. Each connected input is one track.

Other utilities

Sample Objects

Picks a random handful of items out of a list, of any media type. The same seed picks the same items; leave the seed at -1 for a new pick on every run.

Related

• Find and run a workflow
• Apps
• Choose a model
• Reference sets: the source of truth for your art pipeline
• Creative Units