Fields
A field is a living formula — it has a value at every point in space, worked out on the spot, not stored anywhere.
A field is a recipe, not a thing
Most data in DNA is stored. A collection holds a fixed list of rows. A raster holds a fixed grid of pixels. A field is different: it holds no values at all. Instead it holds a rule for producing a value anywhere you ask.
Think of a field as the answer to a question like "how loud is the noise here?" or "what colour belongs at this spot?". You can ask that question at a single point, at a million points, or across a whole surface — and DNA only does the work for the points you actually use.
Because a field is a recipe, it costs almost nothing to carry around and stays infinitely sharp. Zoom in forever and a noise field never goes blocky, because there are no pixels to run out of — the value is recomputed for wherever you're looking.
If you ever want a field as concrete, stored data — to paint it onto pixels, export it, or hand it to something that needs a grid — bake it into a Rasters (images) or a Distance fields & volumes. Baking freezes the recipe into stored values.
The flavours of field
A field always returns something at each point, but what it returns depends on its flavour:
Scalar — a single number at every point. Great for density, height, displacement strength, a mask, or "how much" of anything.
Vector — a direction-and-length (an arrow) at every point. Use it to push, flow, or steer things — wind, currents, forces.
Colour — an RGBA colour at every point. This is how you texture and paint without ever unwrapping UVs. Noise, gradients, checkerboards, voronoi cells, and sampled images all arrive as colour fields.
Distance — at every point it tells you how far you are from a shape's surface (and whether you're inside or out). This is the maths behind smooth blends, rounding, hollowing, and the Analytic shapes pipeline.
Multi-channel (record) — many channels bundled into one field. A material field, for instance, can carry colour, roughness, metalness, normal, and more, all sampled together at the same point.
Morph — a smooth blend between two different things. A morph field can melt one shape into another — a curve into a volume, a mesh into a blob — evaluated continuously so the in-between states are real, not faked.
Different flavours mix gracefully. Ask a colour field for a single number and you get its brightness; ask a scalar field for a colour and you get greyscale. DNA converts automatically so wiring rarely fights you.
Where fields come from, where they go
Fields are produced by generator nodes — noise, gradients (Scene values), patterns, a mesh.sphere's distance, a sampled image — and by the The Expression language node, where you can write your own per-point formula.
Once you have a field, you feed it into something that samples it:
Drop a colour field onto a shape's fill and it paints across the surface, point by point.
Wire a scalar field into a displacement and every vertex looks up its own value to move by.
Feed a vector field into a simulation and particles read the arrow under them and drift along it.
Combine distance fields with booleans to carve, merge, and smooth solid forms.
Because the recipe travels with the field, the same noise can drive colour on one node and displacement on another, in perfect lock-step — they're both asking the same question, just expecting different answers.
Field vs collection vs raster
These three are easy to confuse, so here's the short version:
A collection is stored rows — a definite list of points, shapes, or geometry you can count and edit. See Collections.
A raster is stored pixels — a fixed grid, resolution and all. See Rasters (images).
A field is a stored formula — no rows, no pixels, just a value waiting to be asked for at any point.
Reach for a field whenever you want something continuous, resolution-free, and cheap to carry; reach for a raster or collection when you need concrete, countable, exportable data.