Video Maze V2

Here's a fun one: Video Maze V2 doesn't touch a model at all. It renders a maze - a 3D, first-person, moving-through-corridors maze - straight to frames, using nothing but CPU math and a seed. Wire its images output into a Video Combine and you've got a looping maze flythrough clip with no sampling, no VRAM, no model download. Perfect for background footage, music video B-roll, or that synthwave aesthetic you've been circling.

This is the kind of node that makes DJZ-Nodes feel like a toy box. It's not going to be in every workflow, but it's a great example of "procedural video inside ComfyUI" and it's genuinely fun to tweak.

How it works

The maze itself is generated with a classic recursive backtracker - the algorithm that carves out a perfect maze where every cell is reachable. The twist is that it's seeded, so seed gives you a deterministic, reproducible maze: same seed, same layout, every time.

Then each frame is rendered with raycasting, the same trick old-school 3D engines (think Wolfenstein 3D) used. For every column of pixels, a ray is cast through the maze until it hits a wall; the wall's distance determines its height and lighting. That's why it's CPU-fast - no polygons, no GPU - and why the walls have that authentic retro-3D feel. V2 improves on V1's rendering with adjustable wall_thickness, camera_height, and camera_pitch (tilt the camera up to +45° for a skewed, stylish look, down to -45° for more floor).

The inputs that matter

• width/height (128–4096), fps (1–60), max_frames (default 300) - output geometry and length. That 300-frame default at 30fps is a 10-second loop.
• seed - the whole maze layout. Change it and the corridors reshape.
• maze_size (15–100, default 30) - grid density. Bigger = longer corridors and a grander scale; small = tighter, more frantic passages.
• movement_speed (0.01–0.2) and rotation_speed (0.01–0.15) - how fast you fly and turn. The default is a slow drift; crank rotation_speed for a swooping, cinematic tour.
• fog_distance (5–50) and fov (30–120) - atmosphere and lens width. Shorter fog with a wide FOV is the classic synthwave tunnel look.
• color_scheme - six palettes (neon, classic, rainbow, monochrome, sunset, cyberpunk); wall_pattern - solid, gradient, brick, or circuit; lighting_mode - dynamic, static, pulsing, ambient; render_quality - standard/high/ultra (a quality/CPU tradeoff).
• ceiling_color and floor_color - hex colors for the void above and below the walls.

Output is images (IMAGE batch), frame after frame of your maze tour.

Installing it

Standard DJZ-Nodes install:

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

Restart ComfyUI, or use ComfyUI Manager and search "DJZ-Nodes". Note the requirements include opencv-python, which the maze renderer uses for its line drawing.

Troubleshooting

• It's slow. render_quality at "ultra" with a large maze_size on a big resolution is a lot of raycasts per frame. Drop quality, reduce maze_size, or render smaller and upscale later.
• The camera just drifts into walls. Collision detection is basic; high movement_speed can push you through walls. Keep speed modest (0.05–0.08) and it behaves.
• "Why is my maze the same every time?" That's the seed working as intended. To get variety, vary the seed - hook it to a seed generator node.
• Walls look aliased / jagged. Raise render_quality - the higher modes step the raycast more finely per column.

It's not "real" AI video and it never pretends to be - but as a procedural background generator it's reliable, deterministic, and costs nothing to run. For the right aesthetic, that's a win.