Dinsky Plus Generator V2
Coordinate-Anchored Abstract Art
- IMAGE
Dinsky Plus V2 is the "actually, I want some control" sequel to Dinsky Plus. The original scatters shapes randomly across the canvas; this one lets you pin shapes to specific coordinate pairs. You type in x_coords and y_coords as comma-separated lists, and the node draws circles at those points, with rectangles and lines connecting consecutive points. Same abstract Kandinsky-family aesthetic, but now the composition is yours.
That's a meaningful upgrade if you're using these for anything deliberate - a poster background where you know where the subject will sit, an abstract composition that needs to leave a specific area empty, or a layout where shapes need to relate to each other. Random abstraction is fun; abstraction that leaves room for your text or subject is useful.
How it works
The core is identical to V1: seeded random, PIL drawing, weighted shape selection from the same five palettes, single IMAGE output. The difference is the anchor points. The node parses x_coords and y_coords (commas required; spaces are tolerated), zips them into pairs, and each pair becomes the anchor for one shape. Circles are drawn centered on a pair with shape_radius; rectangles and lines use the next pair as their second point - so consecutive pairs chain together visually. If the parse fails or you give it fewer than two pairs, it falls back to default points at quarter/center/three-quarter positions, so the node degrades gracefully instead of erroring.
The inputs that matter
x_coords/y_coords- comma-separated integer lists, e.g.100,200,300,400,500. Equal-length lists get paired up; the shorter one wins.shape_radius- circle size (1–500).line_width- line thickness (1–50).seed- reproducibility.circle_weight/rectangle_weight/line_weight- shape-type probability mix, as in V1.
Output is a single IMAGE at your requested width/height.
Installing it
Dinsky Plus V2 ships in DJZ-Nodes:
cd ComfyUI/custom_nodes
git clone https://github.com/MushroomFleet/DJZ-Nodes
cd DJZ-Nodes
pip install -r requirements.txt
Restart ComfyUI or install via Manager ("DJZ-Nodes"). Pillow + numpy only - nothing heavy.
Common issues
The coordinates are plain canvas pixels, not normalized - so if you change width/height after laying out your coordinates, everything shifts relative to the canvas. Treat coordinates as part of your design, not an afterthought. Second, the chaining behavior for rectangles and lines means the last pair wraps around to the first (it uses modulo), so an odd count can draw a closing edge you didn't expect. And as with V1, the canvas is always white - no background option, so plan your composite accordingly.
Inputs (11)
| Name | Type | Default | Description |
|---|---|---|---|
| width | INT | 102464–4096 | — |
| height | INT | 102464–4096 | — |
| seed | INT | 00–4294967295 | — |
| x_coords | STRING | 100,200,300,400,500 | — |
| y_coords | STRING | 100,200,300,400,500 | — |
| shape_radius | INT | 501–500 | — |
| line_width | INT | 51–50 | — |
| color_palette | COMBO | 5 options: kandinsky, warm, cool, monochrome, vibrant | |
| circle_weight | FLOAT | 1.00–10 | — |
| rectangle_weight | FLOAT | 1.00–10 | — |
| line_weight | FLOAT | 1.00–10 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| IMAGE | IMAGE | — |