Derp Latent

Derp Latent is the pack's empty-latent generator, and it's the node the author calls "the one that starts it all" - no models, no checkpoints, just a blank canvas in the exact resolution your model wants. Part of xcp_derp-UI (search "xcp_derp-UI" in ComfyUI Manager, by lordwedggie), under derpNodes/ControlDeck.

What it does

Core ComfyUI ships an Empty Latent Image node that does roughly the same thing: you type width, height, batch size, and out comes a LATENT tensor. Derp Latent exists to make that less tedious and more portable. Three interactions carry the day:

• Portrait/landscape toggle - click to flip width and height. Sounds dumb until you've typed 1024x576 backwards for the third time today.
• Resolution selector - click the resolution display to cycle through preset sizes, so you're not typing 512 and 768 forty times a day.
• Batch size - click the number to edit it (min 1, max "however much VRAM you're willing to sacrifice").

Under a settings panel you get Latent configuration, where you define your own resolution presets (custom width/height pairs per model family), and Load Settings, so you can keep a separate profile for SD1.5, SDXL, Flux, Krea2 - whatever resolution targets each model expects, without remembering them. The Krea2 profile even ships as the spawn default, with native 1MP buckets per aspect ratio and reordered ratio lists.

How it works - and the wireless bit

The node's single latent output (type LATENT) is a pure virtual output: the Python shell returns None, and the actual latent is reconstructed from metadata by the pack's signal engine at execution time. The docs spell out that it "broadcasts its output as a wireless signal, you'll need a derpRouter to route it." So yes - this is another node where the Router requirement is real. Add a derpRouter to the workflow or your freshly generated blank canvas goes nowhere.

Two newer toggles in the system panel are worth knowing because they explain why this node beats the core Empty Latent for derp workflows: Output Resolution (default off) broadcasts the chosen width/height as wireless Width and Height INT signals, and Output Latent (default on) gates the wireless latent signal itself. Turn on the resolution broadcast and one node feeds not just the sampler's latent but anything else that needs to know the canvas size - filename templates, resolution-aware logic - without a single extra wire.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/lordwedggie/xcp_derp-UI-Release

Restart ComfyUI. No pip installs, no model downloads, no npm step - the pack vendors its JS. Manager users: search xcp_derp-UI.

Where people get burned

The obvious one: you wire the latent output into a KSampler, hit queue, and nothing happens. The output is a wireless signal; it needs a Router in the graph. Second gotcha: if you have Output Latent toggled off (say you only wanted the resolution broadcast), you get no latent - which is fine if you meant it, confusing if you didn't. And as with every node in this young, barely-documented pack, when behavior surprises you, check derp_docs/Loader Nodes/Derp Latent.md in the repo - the bundled docs are genuinely the best reference that exists.