marduk191 workflow settings

This is the node the whole pack is named after. marduk191 workflow settings is a single control panel that spits out seed, resolution, steps, CFG, denoise, scale factor, sampler, and scheduler - thirteen values in one place, ready to wire into the rest of your graph. If you've downloaded one of marduk191's popular CivitAI workflows (his Flux collection has thousands of downloads), this is the node you'll actually be touching: the "settings bar" his graphs are built around. Instead of hunting across the canvas for scattered widgets, everything you'd normally change between runs sits in one box.

How it works

Under the hood it's a selector plus a lookup table. The Aspect_Ratio dropdown is backed by a 40-entry list covering SD 1.5 (512-based), SD 2 (768), SVD/SV3D, SDXL (1024 up to 2048-wide), and Flux-friendly buckets - the dimensions trace back to the Stable Diffusion aspect-ratio bucket paper (arXiv 2307.01952) and Stability's generative-models repo, so they're the "good" resolutions, not arbitrary numbers. rotation flips landscape/portrait by swapping width and height.

The mode switch is the seed behavior: in Fixed mode it uses your fixed_seed (the default is 8008135, which is... a choice). In Random mode it ignores fixed_seed entirely and re-rolls a new seed on every run, which is how you get fresh results without touching anything.

The inputs that matter

• mode - Random or Fixed. Random re-rolls the seed each queue; Fixed respects fixed_seed. Pick once and stop thinking about it.
• Aspect_Ratio - the big dropdown. This is where most of the value is: all the standard SDXL/Flux resolutions, labeled by ratio, no need to remember what "16:9" should be in pixels.
• rotation - landscape or portrait; handy shortcut that swaps the chosen dimensions.
• The optional block (batch, Pass_1_steps, Pass_2_steps, Pass_1_CFG, Pass_2_CFG, Pass_2_denoise, scale_factor, sampler, scheduler) all have sane defaults, so you only touch what you need.

The two-pass fields give away the intended workflow: Pass_1_* and Pass_2_* steps/CFG plus Pass_2_denoise and scale_factor are laid out for the classic two-pass (hi-res fix) setup - render at base resolution, scale the latent up by scale_factor (default 1.5), then run a second pass at low denoise (default 0.5) to add detail. That matches the standard advice: second pass at fewer steps and 0.3–0.5 denoise. For a single-pass graph, just wire the Pass_1 outputs and ignore the second set - unused outputs cost you nothing.

Wiring the outputs

The outputs are the interesting part because they're typed:

• SEED (INT) feeds a KSampler's seed; TEXT is the same seed as a string, handy if you want it in a display or a prompt token.
• WIDTH/HEIGHT (INT) feed Empty Latent Image (or the pack's own random latent node).
• BATCH_SIZE, Pass_1_steps, Pass_2_steps, Pass_1_CFG, Pass_2_CFG, Pass_2_denoise, SCALE go to samplers and upscale steps.
• SAMPLER and SCHEDULER outputs are typed as the exact enum lists ComfyUI's KSampler expects - so they plug straight into its sampler/scheduler sockets with no string-mismatch headaches.

Install

Zero Python dependencies (empty requirements.txt), no model downloads. Via ComfyUI Manager, search "marduk191" and install this pack, then restart:

1. Manager → Install Custom Nodes → search "marduk191" → Install → Restart.

Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/marduk191/comfyui-marnodes

then restart ComfyUI.

Gotchas

• In Random mode, fixed_seed is ignored. Don't set it and then wonder why your seed keeps changing - that's the mode working as intended.
• TEXT is the seed, not your prompt. The STRING output is str(seed) for convenience, so don't wire it where a prompt goes by accident.
• You don't need all thirteen outputs. It looks intimidating; wire only what your graph uses.
• It's opinionated. Some people prefer scattered widgets and will find a settings node redundant. If you're building your own graphs from scratch, that's fair. But for a shared, reusable workflow - especially a two-pass one - having every knob in one box genuinely simplifies things.

It's not flashy, but it's the reason people who use marduk191's workflows keep this pack installed.