Skip to documentation
AgentDocksDocumentation

Node reference

Image tools

Nine nodes that work on a picture you already have: make it bigger, cut it out, reshape it, describe it, mark it up, or turn it into a layout.

Last reviewed October 25, 2026

Tools work on what exists

A generator is told what to make. A tool is handed a picture and asked to do one thing to it. The image tools sit in the Tools group of the palette, take one image on their input, and give one result on their output. Most have no prompt, because the picture is the instruction.

They come in two kinds. Five are free and run on your device, with no Run button at all. Four spend credits, either because a model does the work or because an AI writes the words.

ToolTakesGivesPrice
UpscaleImageOutput ImageCredits, by model
Remove BackgroundImageOutput ImageCredits, by model
Image DescriberImagePromptCredits, per description
Prompt EnhanceDraftPromptCredits, per pass
Crop & ResizeImageOutput ImageFree
PainterImageOutput ImageFree
ColorNothingText, SwatchFree
Color PaletteExtract fromText, SwatchFree
Design StudioBackgroundDesignFree

The free tools have no Run button

Crop & Resize, Painter, Color, Color Palette and Design Studio are lenses. What you set on the node is a recipe, and the output wire means the picture with that recipe applied. You never press Run on them. When a node downstream runs, the editor bakes the processed picture first, at full resolution, and hands it on. You can also download the baked result from the node.

Edit without paying twice

Because the recipe is applied only when something downstream runs, you can adjust a crop or a paint stroke as many times as you like and pay nothing until the next paid generation reads the result.

Image Describer and Prompt Enhance

Image Describer
Looks at a picture and writes one generation-ready prompt that would recreate it. The text stays editable on the node, and the Prompt output carries your edited version, not the model's raw words. If the picture comes through a free tool, the baked version is what gets described.
Prompt Enhance
Turns a rough draft into a detailed prompt in one pass. Both boxes stay visible: your draft on top, the enhanced text below, editable. With the target on Auto, the node reads the wire going out and tunes the prompt for the model at the other end.

Both write on the text lane rather than the run lane: an action button on the node starts the work, the words stream in, and the Prompt output feeds any node with a text port.

Crop & Resize and Painter

Crop & Resize
The crop box picks what is kept; the width and height fields pick the size it leaves at. Ratio presets sit in a dropdown, quarter turns and flips are exact, and stretching is only possible when the size fields are unlinked.
Painter
Draws circles, arrows, scribbles and short labels on the picture, inside the preview. Feed the marked-up result to an edit-capable model to say which part you mean and what should happen to it. Strokes are stored as vectors and baked at full size.

Color and Color Palette

Color
One exact color from a full picker, with two outputs: Text carries it as a line of prompt text, Swatch carries it as a 600 by 600 image. Saved colors are shared by every Color node in your account.
Color Palette
Up to ten colors, chosen by hand or extracted from an uploaded or wired picture. Extraction runs on your device and the picture never leaves it. The same two outputs as Color, with every chip in the palette.

Two outputs, two uses

Wire Text into a prompt or a prompt variable when the model should read the color. Wire Swatch into a reference slot when the model should see it.

Design Studio

Design Studio is a layered design canvas inside the node: text, shapes and images composed into a finished layout. Quick edits happen on the node itself; the full-screen studio opens for real work. A wired picture becomes the background, and without one the canvas starts from a preset. Every gesture is one undo step on the graph, and the composite is baked before any downstream run reads it.

Common connections

  • Image Preview → Upscale → Image Preview

    Finish a chosen result at print size.

  • Image Input → Remove Background → Image Generate

    Cut a product out, then place it in a generated scene as a reference.

  • Image Preview → Image Describer → Image Generate prompt

    Recreate the look of a result with a different product.

  • Text Input → Prompt Enhance → Video Generate prompt

    Write a rough idea and let the enhancer tune it for the video model.

  • Image Input → Crop & Resize → Image Generate

    Reframe a reference to the ratio the generation needs.

  • Painter → an edit model

    Circle the detail to change and label what it should become.

  • Color Palette Text → prompt variable

    Keep a brand palette in the prompt of every generation.

How references and prompt variables work on Image Generate