CreateSeedNode ♾️Mixlab

The seed is the number that makes a generation reproducible - same seed, same prompt, same everything, same image. In a big graph the pain is when you've got two or three samplers and you want them all on the same seed, or you want one knob to change the seed for the whole workflow instead of hunting through every KSampler. CreateSeedNode (class Seed_) is that one knob: a standalone seed source you set once and wire into everything that needs a seed.

It comes from the "Experiment" corner of shadowcz007's mixlab pack, and it's tied to the pack's headline feature - turning a workflow into a web app (AppInfo). The README notes the pack has a unified seed control that surfaces for SamplerCustom and KSampler, and this is the node behind that. When you publish a workflow as a mixlab app, this seed becomes a first-class control in the app UI, so an end user gets a single "seed" field instead of the raw node graph.

How it works

There's no magic - it holds an integer and outputs it. The value is the whole point. You set the seed here, and every sampler you connect it to draws from the same number, so a "randomize the seed" action or a fixed seed applies across the graph in one move rather than per-node. In app mode it's the node the framework knows to expose as the seed widget.

The inputs and outputs that matter

• seed (INT, default 0, range 0 up to the usual 64-bit ceiling) - the seed value. This is the one thing you set.
• Output seed (INT) - wire it into your KSampler / SamplerCustom seed inputs.

Note it's flagged as an output node, which is a little unusual for something that also feeds downstream - that's so the app framework can treat it as a surfaced control, not a hint that it's a terminal. In a normal graph you use its seed output like any other value.

How to install it

Via ComfyUI Manager: search comfyui-mixlab-nodes, install, restart. Or clone the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/shadowcz007/comfyui-mixlab-nodes.git

then pip install -r requirements.txt (or install.bat on the Windows portable build) and restart. Nothing to download.

Common issues & troubleshooting

My results aren't reproducible even with a fixed seed. A seed only pins the randomness - everything else has to match too. If the model, prompt, sampler, scheduler, steps, or CFG changed, the same seed gives a different image and that's correct behavior. Lock the rest of the graph before blaming the seed.

"control_after_generate" behavior differs from the native widget. The stock KSampler seed widget has that fixed/increment/randomize dropdown baked in. A standalone seed node is a plain value source; how it advances between runs depends on how you drive it (and, in app mode, on the app's seed control). If you specifically want the native increment-after-run behavior, you may still prefer the sampler's own widget - this node's advantage is sharing one seed across many nodes, not replicating that dropdown.

Do I even need it? If you've got a single sampler, honestly no - the built-in seed widget is fine. Reach for CreateSeedNode when you have multiple samplers that must share a seed, or when you're publishing the workflow as a mixlab app and want the seed to show up as a clean control. For a plain one-sampler graph it's extra wiring for no gain.