Text Input (Custom Selector)

If you've ever deleted a wire from your CLIP Text Encode just to try a scene without the "35mm film" line, this node is for you. CustomTextInputSelector_Standalone - display name Text Input (Custom Selector) - takes a few text blocks, lets you pick which ones count with a tiny string like 0,2 or 1:, and hands back both a list and one joined string.

No pixels, no model - it holds a value on one wire and makes the composition of that value editable.

How the blocks stack up

Under the hood it builds one flat list: every prefix block, then text, then every suffix block. The selector is parsed into indices, and those blocks come back out.

The two optional inputs are typed *, ComfyUI's wildcard socket, so they take a string or a list of strings: one string is one block, a list is one block per item. Unwired, they contribute nothing - an unconnected optional input never arrives, which is the None-means-absent convention every switch node in ComfyUI is built on.

So your index is only stable if the block count upstream is stable. Wire a three-item list into prefix and your main text is now block 3, not block 0. Nothing errors; you just get a different prompt.

The inputs that matter

text (multiline) is the main block, and it's always a block: the default is an empty string, so an empty box still occupies an index.

selector (defaults to :) is the actual control surface:

| You write | You get | |---|---| | (empty) | all blocks | | 2 | block 2; -1 is the last | | 0,2 | blocks 0 and 2, in that order | | 1:3 | blocks 1 and 2 (end exclusive) | | -2: | the last two |

Two outputs: text_list, the selected blocks as a LIST, and concat_text, the same blocks joined with no separator at all. concat_text is the one you'll actually use - wire it into any STRING input, most obviously the CLIP Text Encode prompt. Since its text is a widget, right-click it and pick Convert widget to input first - the standard ComfyUI gesture for turning a typed field into a socket.

Installing it

It's a one-node pack from h1531095, first released August 2026, MIT-licensed. Search Text Input Selector in ComfyUI Manager (it's on the Comfy Registry, so it should be indexed), or do it by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/h1531095/Comfyui-TextInputSelector

Restart ComfyUI; it appears under custom/text. Zero dependencies - the pyproject.toml declares an empty dependency list and the README agrees - the best thing you can say about a pack in ComfyUI's world: it can't join the dependency pileup. No model files, no compile step.

Where people get burned

The failure mode here is silence, not a red error box.

• A typo'd selector gives you an empty prompt. Range endpoints get clamped, and a part that isn't a number is swallowed by a try/except. So 1;3 selects nothing, concat_text comes out "", and your CLIP Text Encode happily encodes the empty string. No error anywhere - just an image that ignores your prompt.
• No separator in the join. "B" and "C" concatenate to "BC". Put the spaces and commas in the block text itself.
• Reversed ranges are empty, and there's no step. 3:1 returns nothing. Despite the "Python-slice-style" billing, 0:6:2 behaves as 0:6.
• text_list is a LIST, not a per-item execution. One list value on one wire: it won't run the downstream node per block, and it won't connect to a STRING socket.
• * accepts things that aren't text. Wire an IMAGE or MODEL into prefix and concat_text stringifies its Python repr into your prompt. Strings only, really.

Should you reach for it

It's for one habit: three or four prompt modules - a character line, a style line, a camera line - on fixed wires, dropped or reordered by editing one short field instead of dragging wires. Subject-then-camera-then-lighting is still the ordering that works.

Getting several blocks into prefix needs an upstream node that emits a LIST, because one socket takes one wire. And it's a first-release, single-maintainer pack with no community footprint - ComfyUI's ecosystem has no code review, so read the eight lines of Python if you care. It's dependency-free and deterministic, though, so the blast radius is small. On Flux/Qwen-era LLM-template encoders, remember that prompt weights get discarded and the negative box does nothing. Block selection still works, because it's text assembled before the encoder sees it.