Chaos Randomizer

The Endless randomizer family exists because "everything random" is genuinely fun for exploration, and Randomizer_Chaos is the aspect-ratio-aware one. Feed it min/max ranges for steps, CFG, guidance, dimensions, and seed, and it rolls a full parameter set each run - steps, cfg, guidance, height, width, seed - ready to plug into a KSampler and an Empty Latent Image. The README calls Chaos the node that "randomly changes the aspect ratio on you," and that's its personality: it doesn't just randomize dimensions, it picks a real aspect ratio.

There are three parameter randomizers in the pack: Mayhem (random dims, optionally flipping portrait/landscape), Chaos (aspect-ratio aware), and Pandemonium (a black box with no inputs at all). This is the one to reach for when you want varied compositions, not just varied numbers.

Inputs and outputs

The inputs are all min/max pairs plus a couple of flags - every one has a sensible default, so you can run it untouched:

• steps_min / steps_max, cfg_min / cfg_max - sampling ranges (CFG for SD-style models).
• guidance_min / guidance_max - the separate Flux guidance range, 0-6. That's the cfg_guidance output.
• dimension_min / dimension_max - the size pool the aspect ratios are picked from.
• orientation - square, portrait, landscape, or random (default). Square keeps 1:1; the others draw from a baked-in list of ratios (4:3, 3:2, 16:9, 21:9, 2.39:1 anamorphic, and more, plus inverses). If no ratio fits, it reverts to square.
• divisible_by_64 - off means dimensions snap to multiples of 16; on, multiples of 64.
• seed_min / seed_max / seed - the random range and the seed that reproduces a given roll.

Outputs: steps (INT), cfg_scale (FLOAT), cfg_guidance (FLOAT), height (INT), width (INT), seed (INT). Wire steps/cfg/seed to the KSampler, height/width to the latent node.

Where people get burned

The sanity checks are actually a feature: if you put steps_min=10, steps_max=3, the node silently flips them rather than crashing. The subtle trap is guidance vs. CFG - on a Flux workflow use the guidance output and ignore cfg_scale; on SD the reverse. And a fixed seed input reproduces the whole roll, which is what you want when you finally generate something good and need to rebuild it.

Installing it

Part of Endless 🌊✨ Nodes by tusharbhutt:

cd ComfyUI/custom_nodes
git clone https://github.com/tusharbhutt/Endless-Nodes
# restart ComfyUI

Or ComfyUI Manager → search "Endless". No dependencies or model files.

Gotchas

Beyond the guidance/CFG confusion: dimensions respect your divisor, so don't expect arbitrary sizes; and the pack is a single-maintainer hobby project with a mid-2025 rewrite, so treat it as a well-tested personal tool, not a supported platform. For pure parameter randomization without aspect-ratio logic, the Mayhem node is the leaner alternative in the same menu.