Schematic Overlay

You know the look: a photo drained into white-on-dark, then buried under detection circles, dashed crosshairs, connection lines, coordinate labels and little pixelated zones - like a page torn out of a technical manual. ComfyUI can prompt that aesthetic into a generation, but it comes out half-baked, with gibberish labels and wobbly geometry. Schematic Overlay is the deterministic route: it takes an image you already have and draws the whole annotation apparatus over it in one node. No model, no API key, nothing to download besides the pack - it runs on Pillow, which is CPU-only and fast.

This is a post-processing node, not a generator. It sits at the end of a workflow, after VAE decode: feed it the render (or photo) you actually like, and it stylizes it into "technical data-art" while never touching the pixels outside its own overlay marks. If you want product renders to read as engineering documentation, or a logo shot as a "strategy diagram," this is the fastest route that doesn't involve retraining anything.

How it works

The engine is a grid analysis. The image is split into block_size squares (default 16px), each block gets a score, and blocks above threshold become circle candidates. The detection_mode picks the scoring: combined (the default) multiplies contrast by brightness deviation; contrast, bright, and dark each score a single measure. max_circles caps how many get placed and min_distance enforces spacing so they don't cluster - the whole thing is the classic "detected features on a technical plate" illusion. Circle size is driven by the detection score (between min_radius and max_radius) plus a seeded jitter, so seed changes the distribution of sizes without moving the centers.

Then the ornamentation layers stack on. connection_dist draws straight lines between every pair of circles within that range (set it to 0 to hide lines). The chain_* inputs draw a tangent chain of circles radiating from the canvas center with a shrinking or growing radius per step, plus optional intersection markers with coordinate labels. The crosshair_* group is the dashed full-canvas cross, the dashed centered square frame, and a center asterisk. Pixelation zones get placed either from a coordinate string (pixelate_zones as "x,y;x,y") or from a mask - each connected white region's centroid becomes a zone center. An optional texture image is screen-blended over the whole canvas (if you leave it unplugged and texture_opacity is above zero, a seeded procedural grain fills in), and the four frame_text_* corners give you the studio-slate label look.

The inputs that matter

• image - the source photo. Required, and it's the only input you truly can't live without.
• detection_mode, threshold, max_circles, min_distance - the four knobs that decide whether the circles look like a sparse, confident diagram or a dense cloud.
• seed - reseed the circle-size jitter and the grain fallback. Deterministic per seed.
• palette - four built-ins (whiteOnDark, blackOnLight, goldOnDark, greenOnDark) plus custom, which turns on the bg_color / stroke_color hex pair. Note that custom only applies when palette is set to custom.
• size_preset - match the input, or cover-fit (crop-to-fill, centered) into fixed presets like Square, Landscape 16:9, or Instagram Story. Handy when you're templating for a social slot.

Outputs and where they wire

Three outputs, all from the pack's shared pipeline:

• image - the full composite: background, photo, pixelation, crosshair, connections, circles, frame text, texture, chain, in that order.
• overlay_only - the same stack with the photo removed, so the vector layer can be composited onto anything else.
• overlay_alpha - a 0-to-1 MASK of everywhere a vector element (crosshair, connections, circles, frame text, chain) was drawn. Handy for downstream masking - say, only grain or darken the annotated areas.

Install

Via ComfyUI Manager (search "ComfyUI Schematic"), or the manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/jeremieLouvaert/ComfyUI-Schematic

Restart ComfyUI and both nodes show up under AKURATE/Schematic. The only declared dependency is Pillow>=9.0, which ComfyUI already ships - so no model downloads and no pip side-show. (Schematic Voronoi in the same pack uses scipy, but scipy ships with ComfyUI too.)

Where people get tripped up

The pack is new (first commit August 2026), so there's no backlog of community war-stories - but the grounded gotchas are worth knowing. Circles coming out too sparse? Drop threshold or raise max_circles. Lines everywhere? That's connection_dist - set it lower, or 0 to disable. Want your custom colors to actually show? palette must be custom. And if you're batching video frames, the texture input uses frame 0 for every frame - a per-frame texture won't animate, so connect one or accept the grain fallback. One genuinely nice touch: the same seed produces the same plate every run, so you can tweak one widget and get exactly one clean change.