Video Cube V1

Video Cube V1 is a tiny software 3D renderer whose entire subject is one spinning cube. It builds the cube's vertices, rotates them in 3D, projects them with a real perspective camera (FOV, distance, the works), and draws each frame in one of three render styles - wireframe, solid, or points. No GPU, no model, no dependencies beyond numpy and cv2. It's the kind of node that looks like a toy until you want a reliable abstract rotating object in your compositing pipeline, at which point it's exactly right.

How it works

The math is textbook 3D graphics, done by hand:

• create_cube_vertices / create_cube_faces - the cube's 8 corners and 6 faces
• rotate_points - rotation about X, Y, and Z using the per-frame rotation_x/y/z values (which are in radians per frame - 0.02–0.03 is a gentle spin, 0.1 is a tumble)
• project_points - perspective projection using fov (30–120) and distance (2–20) to position the camera
• render_frame - draws it as wireframe (edges only - the classic 80s demo look), solid (filled faces with per-face lighting), or points (just vertices - which looks fantastic as abstract motion)
• lighting_mode - basic, ambient, dynamic, or none, tinting faces by their normals

cube_size (0.1–5) sets the cube's scale; color_scheme (rainbow, monochrome, neon, pastel, cyberpunk) sets the palette.

Inputs and outputs

Required: width, height (128–4096, step 64), fps (1–60), max_frames (1–9999), cube_size, distance, fov, rotation_x/y/z, color_scheme, render_style, lighting_mode. Output: images - a max_frames-long IMAGE batch.

Where it sits in a workflow

Same playbook as the corridor and maze: this family of DJZ nodes is at its best as a camera-motion supplier. Render the spinning cube, then feed the frames into img2img with a prompt like "chrome cube on a pedestal" and you get an AI-styled object with mathematically perfect rotation - no struggling to coax a consistent camera move out of a video model. In points mode it's also a great abstract-texture source for compositing and motion backgrounds.

Install

Part of DJZ-Nodes by Drift Johnson. ComfyUI Manager → "DJZ-Nodes", or:

cd ComfyUI/custom_nodes
git clone https://github.com/MushroomFleet/DJZ-Nodes
cd DJZ-Nodes
pip install -r requirements.txt

Restart ComfyUI. numpy/torch/cv2 - nothing extra.

Gotchas

Rotation inputs are per-frame deltas, so the values you set are "how much it turns each frame," not a total. Because it's CPU vector math, 4096² runs at high max_frames will crawl - 512² is the sweet spot for speed, and you can always upscale. And distance interacts with fov: if the cube looks clipped or tiny, adjust distance before FOV, since they fight each other.