Load Multi LoRA (Temporary)

This is the one node in the pack that's worth it even if you never touch a URL. Load Multi LoRA (Temporary) - class LoadTempMultiLoRA - is a LoRA loader where you type every LoRA into a big multiline text box, one per line, mixing remote URLs and local files, each with its own strength. It's the batch mode the standard loader never got around to, and it's why people chain a stack of Load LoRA nodes together: character LoRA at 0.7, style LoRA at 0.5, a regional one on top. This node does the whole stack in one place.

The format

Each line takes one of three forms:

https://example.com/anylora.safetensors
0.4:1.0:https://example.com/anylora.safetensors
0.4:1.0:other:https://example.com/anylora.pt

file:anylocal.safetensors loads from your models/loras folder by name, so you can mix a URL you're testing with the LoRAs you already own. For anything you don't spell out, the node's own strength_model and strength_clip (default 1.0) apply - and you can leave fields blank per line (0.1::https://... sets only the model strength). Lines starting with # are comments, which the author added so you stop forgetting which of your eight LoRAs is the "super landscape" one. Fair.

How it works

Under the hood it's just the standard comfy.sd.load_lora_for_models call chained down the list in order - each line patches the previous result, and you get one final MODEL and CLIP pair out. LoRAs it fetches are cached in memory per node, so re-runs with the same URLs don't re-download. Inputs are model, clip, the multiline ckpt_url, plus ckpt_type and download_split like the rest of the pack.

Where people get burned

Where people get burned:

• The format is strict. Each entry must be a bare URL or file: name, or the strength:strength[:type]:target form - anything else throws a "Not according to format" error with the accepted syntax printed.
• URLs with a port break it. The parser splits on colons, so https://host:8000/lora.safetensors gets mangled. Keep URLs to plain https.
• Order matters. LoRAs stack in the order you list them, so put the dominant one last if you want it on top.
• Comments must start the line with # - inline comments after a URL are treated as part of the URL.

How to install

Install is pack-standard: ComfyUI Manager, search "ComfyUI-TemporaryLoader," or cd ComfyUI/custom_nodes && git clone https://github.com/pkpkTech/ComfyUI-TemporaryLoader, then pip install -r requirements.txt (only requests) and restart. Everything lands under the temporary_loaders category. One honest caveat: with several multi-gigabyte models in one workflow, "in memory, not on disk" is a real trade - but for LoRAs, the whole pack's premise finally costs nothing.