Unsplash Image (mtb)

A quick way to drop a real, non-AI-generated photo into your graph - by keyword, at a given size - without leaving ComfyUI to go download one yourself. Feed it "mountain" and a width/height, and it hands back an IMAGE.

Inputs and outputs

• width / height - the size you want back, up to 8096.
• random_seed - 0–100000, controls which random photo you get for a given keyword (useful if you want a reproducible pick rather than a new random image every run).
• keyword (optional) - what to search for, default "nature".

Output: a single IMAGE.

What it's actually good for

This is a placeholder-image node, not a production-reliable one, and it's worth being upfront about why. There's no API key input anywhere in this node's parameters - meaning it isn't calling Unsplash's official, authenticated photo API, which requires an application key on every request. It's built around the kind of keyless, no-signup random-photo access Unsplash used to offer for exactly this casual use case. That style of anonymous endpoint has been tightened up or pulled back across the industry as providers move toward requiring registered API keys, so treat this node as convenient-when-it-works rather than something to depend on for a production pipeline - if it starts timing out or erroring, that's the likely reason, and it's not something reinstalling the pack will fix.

Where it's genuinely handy: mocking up a graph before you have real reference material, testing an image-processing chain (a ControlNet preprocessor, an upscaler, a style transfer) against some real photo without going and downloading one, or filling a background layer in a compositing workflow where the exact source photo doesn't matter. Don't build a workflow around it running reliably every single queue.

Installing it

ComfyUI Manager: search "MTB Nodes," install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/melMass/comfy_mtb

restart, and if it's missing, run pip install -r requirements.txt from inside the comfy_mtb folder - this node needs an HTTP client library to actually make the request, and that's the kind of dependency Manager's automatic install occasionally misses.

Common issues

If the node returns a broken or blank image, or hangs your queue, that's very likely the underlying free image endpoint being unreliable rather than anything wrong with your install - check your ComfyUI console for the actual HTTP error before assuming it's a local problem. Beyond that, this node has no unusual local dependencies, so the generic "some mtb nodes couldn't be loaded" startup message you might see refers to the pack's optional face-swap/DeepBump extras, not this one. If you need a reliable stock or reference image source in a production workflow, this node is the wrong tool - download your reference images ahead of time and load them with a normal image loader instead.