Text as Image

Text as Image rasterizes the string on the CPU using the macOS text engine (Core Text / CoreGraphics): the text is drawn into a high-resolution bitmap with subpixel positioning and high-quality antialiasing, uploaded once to a Metal texture (with mipmaps), and cached. On the GPU, each frame simply draws that texture onto a quad — positioned, scaled, and rotated in Grafit coordinates (origin at viewport center, shorter edge = [-1, +1], positive rotation = clockwise). The result is written to the node's offscreen texture with premultiplied alpha; near-transparent fringe pixels are discarded to keep edges clean on transparent backgrounds.

The bitmap's resolution adapts to the Size input: the internal render size is 256 px × clamp(Size, 0.5, 4.0), drawn at 2× for Retina. In practice text is sharp across the useful size range, but pushing Size (or Scale) far beyond 4.0 magnifies a capped-resolution bitmap and will soften — this is bitmap text, not a distance-field font.

Rasterized textures are cached by text + font + size bucket + color, so a static label costs nothing per frame; only a change to any of those triggers a re-render.

Font selection does not use ports: the font Family and Style (Light, Regular, Bold, Italic, …) are chosen in the node's Settings tab, listing every font installed on the system. The default is SF Pro. Changing the font re-rasterizes the texture immediately.

Alignment decides how the text block hangs off the X/Y anchor point:

The Width and Height outputs report the measured text-block size in Grafit units for the current string, font, and Size — perfect for sizing a background plate behind the text. When Enable is false (or the string is empty), the output texture is cleared to transparent.

Interactions: Size and Scale both make text bigger, but Size re-rasterizes at higher resolution (crisper, slightly more expensive on change) while Scale just stretches the existing texture (free, ideal for smooth animation). Alignment happens before rotation, so a right-aligned label rotates around its anchored edge-center, not the text center.

• Live data readout — any float output → Text for an instant two-decimal display; or format first with String Combine ("BPM: " + value) for labeled telemetry.
• Song/scene titles — String Splitter / Switch selecting from a set of strings → Text, with LFO → Color alpha for fade-in/fade-out title cards.
• Clock or countdown — Timer → Math Expression → JSON Stringify or String Combine → Text for on-screen time displays.
• Text on a plate — wire this node's Width/Height outputs (via Math, adding padding) into a Rectangle node's Width/Height to auto-size a background card that always fits the label.
• Kinetic typography — LFO → Scale and Rotation for pulsing, swinging words; Audio Beat Detection → Enable to flash words on the beat.
• Keyboard-driven captions — Keyboard → Switch → Text to punch different captions live during a performance.

• Generators never draw directly to the viewport: connect Image to a Canvas node (which is also where opacity, blend modes, and 3D placement live).
• Animate Scale, not Size, for smooth continuous zooms — every distinct Size/text/color combination rasterizes and caches a new texture, so rapidly sweeping Size churns the cache.
• Fast-changing text (counters, timestamps) re-rasterizes on every new string; that is fine for readouts, but avoid feeding it high-frequency noise (smooth or quantize the value first, e.g. with Smooth or Math).
• Font Family and Style live in the Settings tab of the node, not the inspector's parameter list — easy to miss.
• Color is baked into the texture; for per-pixel effects (gradients across text, glows) run the Image output through Gradient-masked compositing, Gaussian Blur, or Invert Image.
• Use the text as a stencil: feed Image into a Canvas node's Mask input to reveal video or generative visuals through the letterforms.
• Newlines typed in the inspector's text box are passed straight to the macOS text engine, giving multi-line blocks.