Nebula Generator

Sometimes you don't want a diffusion model to imagine a nebula - you want a nebula now, at a specific size, that you can tint and twist and seed deterministically. That's what Nebula Generator is: a procedural texture generator that builds a full space scene from Perlin-style noise, layered color stops, and a pile of post-processing toggles. No checkpoint, no VRAM, no model. Just numpy doing the math in the time it takes your graph to even start loading weights.

The core is fBm - fractal Brownian motion, or layered noise. You pick a noise family (perlin, simplex, ridged, billow, turbulence), a number of octaves (layers of increasing frequency), and persistence/lacunarity (how each layer's amplitude and frequency grow), and the node builds a smooth, cloudy scalar field. That field is then colored by mapping values through a five-color gradient - color1 through color5, each with its own RGB sliders, defaulted to a purple-blue nebula palette - or one of the procedural color_modes like rainbow, radial, or plasma. Then the post stack kicks in: hue_shift, saturation_boost, brightness, contrast, gamma, plus toggles for warp, bloom, vignette, chromatic aberration, and even stars (star_density, star_brightness, with optional star_colors and star_twinkle).

The inputs that matter

Realistically you will not touch 40 inputs. The ones that do the heavy lifting:

• width / height - canvas size. 64 to 4096, stepped by 64.
• scale - the zoom of the noise. Smaller values stretch the clouds out into big shapes; this is the single most important aesthetic knob.
• octaves - detail layers. 6 is default; more gets wispier.
• color_mode and the five color1..5 RGB stops - where the palette comes from.
• seed - note it's a float from 0–1, not an integer. Move it a little for a new cloud field.

star_density is a fun one: 0.001 gives a sprinkling, up to 0.01 is a full sky. And star_twinkle is a nice touch that costs nothing.

Where it fits

Think of it as a background generator and a compositing asset factory. A procedural nebula makes a great backdrop for a foreground subject, a depth map input for parallax, a texture for something you're going to distort anyway, or just a fast way to generate a sky to img2img against. Because it's fully seedable, you can lock a specific cloud field and re-render at different sizes with the same seed - the deterministic part of diffusion that diffusion itself doesn't give you.

Install

Part of Steaked-nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/StealthNinja1O1/Steaked-nodes

Restart, or install via ComfyUI Manager ("Steaked-nodes"). Dependencies are torch, numpy, Pillow - all standard.

Common issues

The classic mistake is leaving scale at its tiny default and wondering why everything is one giant blur - that's the stretched-cloud look, and it's intentional; raise scale toward 0.05 for finer detail. If your colors all collapse into one blob, check that your five color stops are actually distinct and that contrast/gamma aren't crushing the range. And if the output looks flat, vignette and chromatic are cheap realism boosters. Stars won't show if star_density is 0 - remember it's a density, not a count.