Node reference
These node types cover every workflow. Ports are typed (text, image, video, audio); only matching types connect. Optional inputs can stay unwired; multiple inputs accept more than one connection.
Inputs
| Node | Outputs | What it does |
|---|---|---|
| Image | image | Brings one of your images into the graph — upload, pick from your Library, or pull from a connected integration. |
| Prompt | text | A block of editable text: a description, concept notes, or instructions for downstream nodes. |
| Video | video | Brings one of your videos into the graph. |
Generators
Every generator spends credits when it runs, priced by the model you pick — see Choosing a model.
| Node | Inputs | Outputs | What it does |
|---|---|---|---|
| Generate Image | Prompt (text) · References (image, optional, multiple) | image | Generates an image from the prompt; reference images carry your product, character, or style into the result. |
| Generate Video | Prompt (text) · Frames (image, optional, multiple) · End frame (image, optional) | video | Generates video from the prompt; wired images serve as first/reference frames, and the end frame is the shot the clip lands on. The model list shows only models that use everything you've wired in — connect a product and a person, and you'll see the models that keep both. Pick the clip length from the lengths the chosen model offers. |
| Generate Text | Input (text, optional) · Images (image, optional, multiple) | text | An LLM step — write briefs, shot lists, or prompts. It's vision-capable: wire images in and it reads them. |
| Generate Audio | Prompt (text) | audio | Generates audio from a description — voiceover, music, or sound effects. |
| Edit Video | Video (video) · Prompt (text) | video | Video-to-video: describes a change to real footage — weather, style, set — while keeping the subject and motion. |
| Upscale Video | Video (video) | video | Sharpens and enhances a clip into a delivery-ready master. |
| Join Clips | Clips (video, multiple) | video | Cuts your clips into one finished reel, in the order you connect them — clips that start and end on each other's frames play as one continuous shot. Sound stays in sync. Up to 12 clips, 10 minutes in all. Free. |
Utilities
| Node | Inputs | Outputs | What it does |
|---|---|---|---|
| Combine Text | Texts (text, multiple) | text | Merges several texts into one, with a configurable separator — great for joining a shared style block onto per-shot beats. |
| Split List | text | text × 12 (Part 1–12) | Splits a numbered list into up to 12 separate outputs — the bridge between one LLM-written shot list and a dozen parallel generators. |
| Frame Grab | Video (video) | image | Extracts the first or last frame of a video — the key to chaining clips into seamless transitions. |
| Social Format | Image (image) | image | Resizes an image to a platform preset — Instagram Square, Portrait, Story, Widescreen, Pinterest Pin — cropping to fill or letterboxing. Deterministic and free: one photo becomes every variant without a fresh generation. |
| Split Grid | image | image × 9 (Tile 1–9) | Turns one grid image — a contact sheet, a 3×3 of product shots, a character in four scenes — into separate tiles, left to right then top to bottom. Pick the shape from 1×2 up to 3×3; generate the grid once, then send every tile on to its own video or edit. Free, like Social Format. |
| Output | Image · Video · Audio · Text (all optional) | — | Marks a deliverable. Wire your finished results here so runs surface them cleanly. |
:::tip Frame Grab is free Frame Grab extracts frames without spending any credits. :::
Keep going
- See these nodes in action: Building workflows.
- Study real graphs that use them all: Workflow templates.