Load Prompt From Folder X Lora (Soze)

If you've ever wanted to test twenty prompts against five LoRAs and gotten as far as "okay, that's a hundred manual runs" before giving up, this is the node that was built for exactly that afternoon. It reads a folder of prompt text files, walks a rotation of LoRAs, and does it all off a single counter - one index value drives which prompt-file-and-LoRA pair you're on, so a queue that auto-increments that counter marches through the full grid on its own.

How it works

Soze's README describes the same trick on the pack's CSV nodes: the X Lora and X Checkpoint variants "iterate index across rows × models, so a single counter walks every (row, checkpoint) pair." This node applies that idea to a folder of prompt files instead of a CSV - the prompt files are your rows, and start_lora_name plus lora_count define the LoRA rotation. Bump index on each queued run and you cover every combination without rewiring anything between runs.

The inputs and outputs that matter

• input_folder - the directory holding your prompt .txt files.
• start_lora_name - a dropdown, populated from whatever's in your models/loras folder, marking where the LoRA rotation begins.
• lora_count - how many LoRAs from that starting point to cycle through (1–100).
• index - the counter that picks your position in the combined sweep. Drive this from ComfyUI's own auto-increment on the queue rather than hand-typing it once you're past a couple of runs.

The outputs split cleanly into two families. For the prompt side you get Prompt (the file's text content), Input_Path, Prompt_Filename_Path, Prompt_Filename, and Prompt_Filename_No_Ext - handy for naming your saved outputs to match the prompt that generated them. For the LoRA side you get Lora_Full_Path, Lora_Name_Only, and Lora_Index. There's also a status string for a quick readout of what the node actually resolved on this run, useful for confirming a batch is progressing the way you expect without opening a preview every time.

Wire Prompt into a CLIPTextEncode. Wire Lora_Full_Path straight into this same pack's own Lora File Loader - it takes a LoRA by absolute path instead of the usual dropdown, so the two nodes were clearly designed to sit next to each other in a batch pipeline.

Installing it

Via ComfyUI Manager, search ComfyUI_Soze or "Quality of Life Nodes for ComfyUI." Manually: cd ComfyUI/custom_nodes && git clone https://github.com/SozeInc/ComfyUI_Soze.git, then pip install -r ComfyUI_Soze/requirements.txt inside your ComfyUI environment, then restart. No model downloads, no GPU-side dependency - it's pure file and string handling.

Worth setting expectations: this isn't a pack with the reputation of something like rgthree or Impact Pack - it started life folding in the now-abandoned Comfy_KepListStuff helpers and reads more like one person's personal batch-testing toolkit that happens to be public. That's fine for what it does, but don't expect a deep well of community troubleshooting threads if something looks off; the GitHub issues page is your best bet.

Common issues

If start_lora_name's dropdown is empty, your models/loras folder is empty or ComfyUI hasn't rescanned it - restart after adding files. If input_folder has no .txt files, keep an eye on status rather than assuming silence means success. And remember index tops out at 1,000,000 but your actual grid is only (prompt count) × lora_count combinations wide - going past the last valid pair is on you to track, the node doesn't wrap or warn.