LoRA Loader With Name Stacker

This is a normal LoRA loader with one extra trick: as it applies a LoRA, it also builds up a running text record of which LoRAs you used and at what strength. That record chains from one loader to the next, so a stack of these nodes leaves you with a single string listing every LoRA in the pipeline - which you can then feed to a metadata/image-saver node so the info actually gets embedded in your output.

Why does that matter? Because ComfyUI is notoriously bad at recording LoRA info in a way other sites can read. The author is blunt about it: this is "for any website that could not identify all LoRAs' information from image. Apparently that only happens with ComfyUI." If you've ever uploaded a ComfyUI image to Civitai and had it fail to detect your LoRAs, this is the workaround. It's part of ComfyUI_Mira, mirabarukaso's toolkit, and it's aimed squarely at people who share their work and want the LoRA credits to travel with the image.

How it works

Functionally it loads the selected LoRA onto your model and clip at the given strengths, exactly like the standard loader. The addition is the lora_stack string: each node appends its own LoRA name and strengths to whatever came in on its optional lora_stack input, passing an ever-growing list forward. Chain several and the last one's lora_stack output holds the full set. If bypass is on or all strengths are 0, it passes the stack through unchanged.

The inputs and outputs that matter

• model / clip - the model and CLIP to apply the LoRA to.
• lora_name - the LoRA to load (a dropdown of your installed files).
• strength_model / strength_clip - the two strengths (-10 to 10, default 1).
• bypass - skip this LoRA without unwiring it.
• lora_stack (optional input) - the running string from a previous loader.
• MODEL / CLIP (outputs) - the patched model and CLIP, into your sampler or the next loader.
• lora_stack (output) - the updated record, ultimately into an image-saver's metadata field.

Installing it

ComfyUI Manager: search ComfyUI_Mira, install, restart. Or clone:

cd ComfyUI/custom_nodes && git clone https://github.com/mirabarukaso/ComfyUI_Mira.git

then restart. If it errors on load, run pip install -r requirements.txt in the ComfyUI_Mira folder. It's under Mira/Lora.

Common issues

• Leave the first one's stack empty. The author's own reminder: "leave the locomotive (first one) empty." Don't wire anything into the first loader's lora_stack input - it starts the chain.
• Second hires pass ignores a repeated name. Per the docs, if you reuse the same LoRA name in a second (hires) pass, it's ignored. Plan your two-pass stacks accordingly.
• The stack is just text. It records names for metadata; it doesn't change the image. You still need a saver node that reads lora_stack and writes it into the file for it to reach Civitai or wherever.
• Strengths and family still apply. This loads a real LoRA - keep it matched to your checkpoint family (SDXL LoRA on an SDXL base, etc.), and remember bypass/zero-strength means it does nothing but pass the record along.