文本处理

This is the biggest node in the pack and the one that's hardest to explain in a sentence, because it's really three nodes in a trench coat: a joiner, a splitter, and a selector. You feed it text (from the text box, or from up to 64 other nodes), tell it how to cut the result up, and then choose which pieces come out on which port.

If you've ever pasted a multi-line script into a prompt field and thought "I want each of these to be its own branch," this is the node for that.

Joining: 64 wildcard inputs

text is the base text box. On top of that, any_1 through any_64 accept anything - text, numbers, lists, which get stringified on the way in (lists join by newline, bytes decode as UTF-8). 输入端口 sets how many of those ports exist, and you have to click 更新端口 on the node for the change to take effect.

When you've connected any of them, the port contents are joined with triple newlines and replace the base text. When you haven't, the base text is used as-is. Order is port order, which is what makes this the standard way to concatenate three text sources from three different places into one paste-able block.

Splitting: eight modes

分段方式 is where the personality is:

• 空行 (default) - splits on double newlines and rejoins within a paragraph. The natural mode for pasted prose.
• 端口 - one paragraph per input port, strictly by port index. With a gap in the middle you get an empty paragraph in that position rather than a shift, which is deliberate: 段落2 always means port 2.
• 序号 - starts a new paragraph at a list marker. The regex is broader than the READMEs usually admit: 1. style counters, bracketed forms like (1), roman numerals, Chinese numerals, and bullet glyphs all count as markers.
• 段落 - every line is its own paragraph. Blunt and predictable, and what you want for line-per-prompt text.
• 标题 - heuristic chapter detection. Short lines (20 characters or fewer, after trimming) that don't end in punctuation get treated as headings; a run of headings accumulates into the block that follows. Works nicely on a structured document, produces nonsense on a casual one.
• 数字 - keep only the numbers. Genuinely useful as a parser for "extract every numeric value from this pile of text," and great for grabbing values out of a filename dump or a parameter listing.
• 地址 - extracts Windows paths (D:\Data\img.png) out of messy strings, including Python-object reprs. That's exactly what it's for: recovering file paths from text that was never meant to be read by a human, ready for 加载批量图像's track JSON or a save-prefix field.
• 手动 - you insert ||| where the cuts go.

段落优化 toggles whitespace hygiene: on (default) trims each line and joins paragraphs with spaces, off preserves the original indentation and line structure. If you're splitting markdown or code, turn it off.

Selecting: which paragraphs come out where

选取段落 takes -1 for everything, empty for nothing, or indexes like 0,2,4 for paragraphs 1, 3 and 5 - and it does not have to be comma-separated: full-width commas, 、, dots and slashes are all accepted as separators.

Two outputs matter, plus the giant bank of per-paragraph ports:

• 数: - an INT with the count of selected paragraphs. Great for driving a loop-ish section of a graph or for sanity-checking your split without previewing anything.
• 总段: - a STRING list of the selected paragraphs. This is the one you usually wire into something downstream.
• 段落1 … 段落100 - individual STRING outputs, generated by setting 输出段落 (click 更新端口 again) and then wiring each one wherever it belongs.

There's one nuance that follows from the author's design rather than conventional expectations: in 分段列表 mode, the 段落x ports stay aligned to the original paragraph index - select paragraph index 2 and it appears on 段落3, with 段落1 and 段落2 empty. Switch 输出模式 off (输出原始文本) and the selected paragraphs instead collapse into one joined string that lands on 段落1. Which means 段落x port alignment changes meaning when you flip 输出模式. Know which mode you're in before you wire a dozen ports.

Install

yanhuo-easy-use, 16 nodes, category yanhuo, Chinese labels, MIT, v1.3.1. ComfyUI Manager → search the pack title → install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/yanhuoxunxianren/yanhuo-easy-use
# restart ComfyUI

No pip install - pyproject.toml declares no dependencies - and no model downloads. It's a string node; it works with no GPU at all. The 64 input and 100 output ports exist in the class definition already, so expanding the port counts doesn't need another file write; it's the frontend that adds the sockets.

Common issues

"My new ports didn't appear." Change 输入端口 or 输出段落, then click 更新端口. The number alone does nothing.

"Everything came out on 段落1." You're in 输出原始文本 mode, where selected paragraphs are joined into the first port. Flip 输出模式 on for per-paragraph output.

"The split landed in the wrong place." The default 空行 mode wants a blank line between paragraphs. Single newlines don't count - that's what 段落 mode is for. And 标题 is a heuristic; if it misfires, you're asking it to guess.

"A whole paragraph vanished." Check 选取段落. A leftover 0,1,2 from a previous experiment silently drops everything else, and that's the most common version of this complaint.

"The node returned nothing at all." Empty input returns zeroes and empty strings rather than erroring, so an empty result usually means the upstream text node produced nothing - or 选取段落 was left blank, which means "select nothing" by design.