💭 PD Wedge

The most useful habit in image generation is also the most tedious: lock the seed, change one variable, compare. Now do it twenty times. 💭 PD Wedge automates exactly that. You give it a name, a type, and a list of values - one per line - and on each execution it emits one value from the list. Enable auto_queue and it keeps going, queueing the remaining iterations against your local ComfyUI server until the whole list has run. It's a focused little iteration engine, and it pairs perfectly with the pack's output nodes: your filenames can carry the wedge value so nothing gets mixed up.

How it works

The node holds a list, tracks a current_index, and on execution returns three things:

• wedge (the WEDGE type) - the full state dict (name, values, current index). This is what you feed to other Pipedream nodes that understand wedges.
• current_value (STRING) - this run's value, cast to your chosen type. Wire it into the parameter you're iterating: a prompt, a seed, a CFG, a sampler choice.
• wedge_index (INT) - which iteration this is.

In sequential mode it walks the list in order. In random mode it shuffles deterministically per index, so every iteration still gets a reproducible draw rather than a fresh random pick.

The headline feature is auto_queue: on the first execution, the node POSTs modified copies of the current prompt to ComfyUI's API - one per remaining value - with the wedge index injected, so the whole list runs back to back. You press Queue once, walk away, and come back to a finished iteration set. It talks to http://127.0.0.1:8188 by default (override with the COMFYUI_SERVER_URL environment variable).

The inputs that matter

• name - the wedge's identifier. This becomes a token: a wedge named prompt is referenced as {wedge_prompt} in output templates, which is how the file naming tracks which iteration produced what.
• type - STRING, INT, or FLOAT. Determines how values are cast; an unparseable value falls back to string with a console warning.
• values - one per line. Empty lines are skipped.
• mode - sequential or random.
• auto_queue - on/off. Off, you step through manually by re-queueing; on, it runs the list unattended.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/TheOnlyAaron/comfyui-pipedream

Restart ComfyUI, or install "Pipedream" via ComfyUI Manager. No dependencies, no model downloads.

Gotchas

• auto_queue needs the ComfyUI API reachable on localhost:8188. If you run ComfyUI behind a reverse proxy, on another port, or remotely, set COMFYUI_SERVER_URL - otherwise the queue silently stops after the first iteration with an error in the console.
• Each wedge iterates on its own. If you connect two wedges into one workflow, they don't automatically combine into a cartesian run - that's what 💭 PD Rasterizer previews and why you check total_iterations before you start.
• The empty-wedge warning ("no values to iterate") is a clue, not a crash: the node emits a harmless empty value. Double-check the values field isn't just whitespace.
• Iterating over seeds is the classic use, but remember iteration discipline: change one variable per list, or your comparisons are worthless no matter how many renders you queue.