Extract Power LoRAs

rgthree's Power Lora Loader is the default way to stack multiple LoRAs in one node - and its only real flaw is that it doesn't tell you, in plain text, which LoRAs it applied. AUNExtractPowerLoras fixes that: it walks your workflow graph, finds the Power Lora Loader nodes (and some other LoRA loaders), reads out every LoRA name and strength, and hands them to you in three different string formats so you can put "which LoRAs were on" into a filename or a sidecar.

The single required input is loras_delimiter - the separator used when joining LoRA entries into the token string. The default is ;, and the tooltip suggests +, -, _, , or ; depending on how you like filenames to read. The choice mostly affects the first output.

Those three outputs are the real story, because they're formatted for three different jobs:

• loras_token - a single grouped token string with entries joined by your delimiter, sanitized so it's safe for filenames. This is the "dump it into the filename" output, e.g. loraA_0.8;loraB_0.6.
• loras_names - delimiter-joined short names with strengths, but no parentheses. Readable and compact; good for a label or a caption line.
• loras_list - newline-separated base names (file stem, extension stripped) with labeled strengths when available, one per line. This is the "give me the list" output for sidecars, logs, or a Show Text node you can read.

The mechanism is graph scraping, the same family of trick as the pack's AUNGraphScraper and AUNExtractModelName: it reads the workflow JSON, finds LoRA loader nodes by class, extracts their lora_name/strength widget values, then runs them through filename sanitization. It's designed to re-evaluate each run (IS_CHANGED forces it), so if your Power Lora Loader slots get toggled between runs, the extracted text follows along. And it leans on shared helpers with the pack's own LoRA nodes, so the short-name output matches what the rest of the AUN filename machinery would produce - consistency you'll appreciate when the token in the filename matches the LoRA you see in the UI.

Where this earns its keep: you've built a workflow around AUNRandomLoraModelOnly or Power Lora Loader with per-run randomization, and you want the save step to record exactly what combination produced the image. Wire loras_token into AUNPathFilenameV2's filename template and you get searchable, self-documenting filenames.

From the AUN pack: ComfyUI Manager (search "AUN") or cd ComfyUI/custom_nodes && git clone https://github.com/loz2754/AUN-ComfyUI-Nodes, then restart. It needs rgthree's Power Lora Loader to be installed to be useful - that's rgthree/rgthree-comfy - but the node itself is dependency-light and needs no models.