Text Loader Crawl Batch (CRT)

Prompt batching is one of those things everyone eventually hacks together - a text file per prompt, a load node, a counter - and Text Loader Crawl Batch (CRT) is the version where the scaffolding is already built. It grabs a batch of .txt files from a folder in one run and hands them to your graph as strings. If you've got a folder of 50 prompts and you want 50 renders without babysitting the UI, this is the node you reach for.

Like the other "Crawl" loaders in CRT-Nodes, it's folder-driven and seed-indexed. The workflow pattern: bump a seed, the node steps through your files, your sampler renders, a save node writes the result. Rinse, repeat, come back to a finished folder.

How it works

The node scans folder_path, filters by file_extension, and sorts the results in natural order - so 2.txt sorts before 10.txt, which is exactly what you want when your files are numbered. Then seed acts as an offset into that sorted list, and it reads batch_count files starting there. Set seed to 0 and it starts from the first file; bump seed by batch_count each run to walk the whole folder.

Its outputs scale with batch_count - one text string per file loaded (the schema lists text_output_1, and more appear as you raise the batch size). Wire those into a positive prompt or a text-conditioning node.

The inputs you actually set

• folder_path - absolute path to the folder. The one input you can't skip.
• batch_count (1–64) - how many files to load per run.
• seed - the batch offset, not a randomness seed. 0 = start at the first file.
• file_extension - default .txt; change it if your prompts live in .md or .prompt files.
• max_words - cap each output at N words (0 = no limit). Handy when you're feeding a prompt budget and want every variant the same length.
• crawl_subfolders - include files in subdirectories or not.

Where it earns its keep

• Long prompt sweeps: hundreds of prompt variants, one node, one seed counter.
• A/B testing: put each variant in its own file, batch them through the same sampler and LoRA stack.
• Multi-prompt batch renders: load several prompts at once into a batched workflow and sample them together.

Gotchas

• Empty folder or a bad path returns empty strings rather than erroring loudly - your KSampler will happily run on a blank prompt, so spot-check the text outputs if results look off.
• The seed is an offset, not an index into "the file I want right now." To review file #7 in isolation you'd set seed to 6 (0-based) with batch_count 1.
• Keep the extensions consistent - a folder mixing .txt and .md files only matches the one you set.

Install is the shared CRT-Nodes story: ComfyUI Manager → search CRT-Nodes → install, or git clone https://github.com/PGCRT/CRT-Nodes into custom_nodes, then restart. It's a big suite with a heavy dependency list, but it degrades gracefully - if an optional import fails, the pack just skips that node instead of taking down ComfyUI.