ComfyUI : XY Plot Tutorial. You will use this a ton!

You've been there. Tweaking one setting. Generating. Squinting. Tweaking another. Generating again. Forty-seven images later you can't remember which one used karras and which one was exponential... and your eyes are crossing. There's a better way.

XY plotting in ComfyUI is one of those tools that makes you wonder why you ever did it the hard way. Instead of generating one image at a time and trying to hold the differences in your head, you generate a single grid that lays everything out side by side. Schedulers across the rows. Seeds across the columns. BAM... visual clarity.

The whole setup lives inside the Efficiency Nodes pack for ComfyUI. If you don't have it installed yet, pop open your ComfyUI Manager, find "Efficiency Nodes for ComfyUI," and get it loaded. This pack does something I usually warn people about... it bundles a lot of functionality into single nodes. But it does it so well that it earns its place.

Building the Workflow

Start with the Eff. Loader SDXL node. This one consolidates your model selection, prompts, dimensions, clip skip, and refiner checkpoint into a single clean block. All the stuff you'd normally wire across five or six nodes? One node. Drop in your SDXL base model, your refiner, a prompt, set your resolution to 1024×1024, and you're rolling.

The loader outputs a tuple... think of it like a packed suitcase containing your model, clip, VAE, and latent. You can unpack it if you need individual pieces, but for today we're passing it straight into the next node.

That next node is the KSampler SDXL (Eff.). Wire the tuple right in. Now here's the detail that will save you twenty minutes of confused troubleshooting... at the top of the KSampler, there's a sampler state dropdown. It has three options: Sample, Hold, and Script.

Set it to Script.

If you leave it on Sample, the node generates one image and stops. You'll stare at your screen wondering why your beautiful XY plot isn't plotting anything. Script mode tells the KSampler to accept instructions from an external script node... which is exactly what we're about to connect.

The XY Plot Itself

Drop in the XY Plot node. This is the conductor of the orchestra. It takes dependencies from your loader and script instructions for what to vary across each axis.

For the X axis, add an XY Input: Seeds++ Batch node. This controls your columns... each column gets a different seed, giving you variation to compare against. Set your batch count to however many seeds you want to see. Three is a solid starting point.

For the Y axis, add an XY Input: Sampler/Scheduler node. This is where it gets interesting. You manually select which schedulers you want to compare... normal, karras, exponential, whatever catches your curiosity. Each one becomes a row in your output grid.

Wire Seeds into X. Wire Scheduler into Y. Wire the XY Plot's script output into the KSampler's script input. Wire your latent and VAE connections from the loader to the KSampler.

That VAE connection matters. Skip it and you'll get a cryptic error that just says "zero." No helpful message. No breadcrumb trail. Just... zero. Now you know.

A Small Gift: Live Preview

Before you hit generate, try this. Open your ComfyUI Manager, find the preview method setting, and switch it from "None" to TAESD (the slow version). Close the menu, and now your KSampler node will show you a live rendering preview as each image builds. It's not necessary, but watching the grid fill in piece by piece? Satisfying.

The Payoff

Hit queue. Grab some water. The generation takes time because you're building a full grid... but when it finishes, you've got a single image that shows you exactly how each scheduler performs across multiple seeds. No guesswork. No flipping between tabs. No trying to remember what you changed three generations ago.

You can flip the grid orientation if you prefer seeds on rows and schedulers on columns. You can adjust spacing. You can save the output directly from the preview node. And here's the beautiful part... if you drag and drop that saved image back into ComfyUI, it loads the entire workflow. Your XY plot. Your settings. Everything.

Why This Matters Beyond the Technical

This isn't really about schedulers. It's about building a system for making better decisions. Systematic comparison beats gut feeling every single time when you're dialing in creative parameters. The same principle applies whether you're comparing schedulers, samplers, CFG scales, or model checkpoints.

The XY plot is a thinking tool disguised as a ComfyUI node. It forces you to ask: What am I actually comparing? What variables matter? What does the evidence show me?

Save your workflow as a node template. Next time you need to compare anything... swap out the Y axis input for whatever parameter you're curious about. Seeds stay on X. Your comparison target goes on Y. The framework stays the same.

One grid. Multiple variables. Clear answers.

Finding that special place where work and play intertwine is magical for creating deep neural connections. XY plotting is exactly that kind of intersection... methodical enough to give you real data, creative enough to surprise you with what you discover. Stop generating blind. Build the grid. Let the evidence show you what your eyes alone can't hold. 🎯

--- Source: https://www.youtube.com/watch?v=gbSvsSpfnJs