Qwen3-VL Prompt List Builder

Where Prompt Splitter hands you captions one at a time by index, Qwen3-VL Prompt List Builder does the opposite: it takes the same batch output and flattens all the captions into a single string, one per line (or separated however you like), plus a count. If your downstream target is a batch image-generation node - something that wants a whole list of prompts in one text field - this is the node that feeds it.

Think of the pipeline. You run Qwen3-VL Batch Run (Folder) over a folder of reference images and get a blob of === filename === blocks. That blob is great for humans reading results top-to-bottom, but it's the wrong shape for a batch sampler, which wants "prompt one, prompt two, prompt three" as a flat list. Prompt List Builder does the reshape: parse the batch output, strip the filename markers, join the remaining captions, done.

The three inputs

• batch_text (STRING, multiline) - the batch output, pasted or wired from Batch Run.
• clean_prefix (BOOLEAN, default true) - strips the Next Scene: prefix from each caption, same as the splitter. On by default; turn it off if your downstream generator actually wants the prefix.
• separator (enum) - how to join the captions: newline (default), |||, or ;;;. This is the knob that matches whatever your downstream node expects. Newline is the safe default for most batch prompt inputs; the custom separators exist because some nodes parse on a specific delimiter.

The two outputs

• prompt_list (STRING) - all captions joined by the separator, filename markers gone. Straight into your batch generation node's prompt field.
• count (INT) - how many prompts ended up in the list. Handy for sanity checks (did the batch really caption all 200 images?) and for configuring a batch sampler's count.

What it doesn't do

Worth being explicit, because the name invites the assumption: this node does not build prompts from scratch, and it does not apply any per-prompt logic. It's a formatter. If you feed it text that isn't in the === filename === batch format, it treats the whole string as one prompt and returns count = 1 - no error, just a uselessly flat list. If you need one caption extracted at a time with its filename, that's Prompt Splitter's job; these two are the split-into-one and flatten-into-many halves of the same parser.

Where it fits

The classic use: batch-caption a folder of reference images with Batch Run, run the output through Prompt List Builder, and hand prompt_list to a batch image-generation node so each caption becomes the prompt for regenerating its source image (or a variant of it). The count output tells you how many generations you're about to kick off, which is a nice last checkpoint before you queue up 200 jobs.

Install once via ComfyUI Manager (search "qwen3-vl-comfy-ui") or cd ComfyUI/custom_nodes && git clone https://github.com/Granddyser/qwen3-vl-comfy-ui && pip install -r requirements.txt, restart, and you'll find it under Qwen3-VL with the rest of the batch helpers. It's a small node - a parser, really - but it's the one that closes the loop between "folder captioned" and "batch generated."