Folder File Read

Folder File Read is File Read Line's bigger sibling: instead of iterating over the lines of one text file, it iterates over the files in a folder, reading one whole file each time you run the graph, then wrapping around when it reaches the end. Point it at a folder of prompt files and queue the graph a few times and you're batch-testing across files with zero re-wiring. It's from the TakkunRed ComfyUI-String-Function pack, and like everything in that pack it's pure Python stdlib - no dependencies, no model downloads.

This is the node for the folder-shaped batch job: a pile of caption files you want to audit one by one, a directory of prompt variants, a set of config snippets for an LLM node. Line-by-line files don't fit that shape; files-in-a-folder do.

How it works

Mechanically it's the same always-rerun state machine as File Read Line: the node keeps an in-memory counter keyed to its own node instance, and its IS_CHANGED returns float("NaN") so ComfyUI re-executes it every run instead of caching it. Each run it returns the file at the current position and advances. sequential mode walks the sorted list and loops; shuffle consumes a random order without repeats and reshuffles when done.

The parts that make it actually pleasant to use:

• extension_filter - default .txt, comma-separated for several (.txt,.md), and blank means all files in the folder. Filtering happens before counting, so file_count reflects only what you'll actually read.
• sort_order - name_asc / name_desc for a stable alphabetical walk, or modified_asc / modified_desc to work through newest-or-oldest first. If you're processing a folder that grows as other nodes write into it, sorting by modified time is the sensible default.
• reset - feed True on a run and it jumps back to the first file that same run.
• The counter is RAM-only. Restart ComfyUI and you start from the top; every node instance keeps its own independent counter.

Inputs and outputs

folder_path is the only required input. Beyond that: mode, reset, extension_filter, encoding (the same auto → utf-8-sig → utf-8 → cp932 chain as File Read), and sort_order.

The outputs are where the file-level iteration pays off:

• text (STRING) - the current file's full contents. This is what feeds your encoder or LLM node.
• file_name (STRING) - just the filename, e.g. prompt_03.txt, great for a filename_prefix on SaveImage so each output is labeled.
• file_path_out (STRING) - the full path, useful if a downstream node wants to write a sibling file or save metadata alongside it.
• file_index (INT) / file_count (INT) - position and total.
• is_last (BOOLEAN) - True on the final file before the loop wraps.

Installing it

Search "ComfyUI-String-Function" in ComfyUI Manager, or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/TakkunRed/ComfyUI-String-Function

Restart ComfyUI and find it under String Function. MIT-licensed, zero third-party deps.

Where people get burned

• Same cache cost as File Read Line. The NaN IS_CHANGED forces a rerun every time, which also forces everything behind it to re-execute. Fine for a light graph; annoying if this sits in front of a heavy sampler - you'll want to isolate the iterative section.
• Counters aren't persistent. Restart, and you're back at the first file.
• The folder picker button is Windows-only. The "📂 フォルダを参照..." button launches a PowerShell dialog on the backend; on Linux or macOS it errors, so type the path.
• One run = one file. Queue the graph N times to process N files in one go - the counter advances once per execution, not once per second.