SK Prompt Matrix Sampler V2

This is the version of the prompt matrix sampler most people actually want to reach for, because it removes the one annoying thing about V1: no wire required. SK Prompt Matrix Sampler V2 is functionally the same sampling node - each line is a dimension, |-separated words inside a line are the options, one option gets picked per line and joined into a single prompt - but the text lives in a built-in multiline widget right on the node.

If you read the V1 article, you already know the mechanics, so here's the delta and nothing more. The one input that changes everything is raw_text: a multiline STRING widget with "" as its default, where you paste your grid. No feeding it from another node, no wondering why typing does nothing. Everything else matches V1 exactly - pick_count (0 = all lines, capped at the count), sort_mode (Sequential or Random), input_sep (default |), output_sep (default , ), and seed for reproducible draws. Same single string output into your CLIPTextEncode.

The practical advantage is speed. For a quick comparison - say, one character LoRA against a style LoRA against both - you can type the grid directly, run it, change the seed, run again, and never touch the rest of the graph. The trade-off is that the grid is now buried inside the node's widget rather than living as data elsewhere in your workflow, so if you want to regenerate the grid from other nodes or share it as part of the graph's logic, V1's wired input is the better shape. For casual matrix testing, the widget wins.

A quick grid looks like this (one dimension per line, options split by |):

photorealistic|cinematic|anime
portrait|full body
golden hour|studio lighting

With pick_count at 0 and Sequential mode, one run might give you photorealistic, portrait, golden hour - and a different seed will draw a different combination from the same lines.

Installing. ComfyUI Manager (search "SK LoRA Manager") or:

cd ComfyUI/custom_nodes
git clone https://github.com/sinvks/ComfyUI-SKLora-Manager.git
pip install -r requirements.txt

Restart and it's there. No model downloads, no extra dependencies.

The usual reminder applies: this returns one sampled prompt per run, not a fan-out of every combination. If you need every permutation emitted at once, this isn't that node - but if you want a deterministic one-shot sampler you can poke with a seed, V2 is the friendliest of the three versions.