Load Multiple LoRAs by name (Online)

Load Multiple LoRAs by name (Online) is the pack's take on a multi-LoRA stacker: instead of chaining three LoraLoader nodes, you type a comma-separated list into one text field and it applies them all. It's the same idea rgthree made famous with Power Lora Loader - compact, one node, multiple adapters - except here the LoRAs come from DoubTech's online catalog rather than your local folder, and the "UI" is a text string.

How the input format works

Two sockets come in - model and clip from your checkpoint loader - and one string field, loras. The format is:

name:strength_model:strength_clip,name2:strength_model:strength_clip

The default value spells it out: disney-princess:1.0:1.0,cyberpunk-animal:.5:1. Rules to know:

• Entries are comma-separated; each is split on colons.
• Strength values are optional - name alone defaults both to 1.0.
• If the name has no extension, .safetensors is appended automatically.
• They're applied in order, left to right, each one building on the last.

Outputs are MODEL and CLIP, which you feed straight into your sampler just like the core node.

The mechanism is the shared pack machinery: the catalog comes from DoubTech's API, each named LoRA is downloaded into models/loras on first use and cached in memory afterwards, then applied via ComfyUI's own load_lora_for_models. No API key, no hosted inference - the files end up on your disk and it's plain local loading from there.

The trap: don't trust the default string

Here's the thing the brief won't tell you: the default LoRA names are not in the current catalog. I checked the live API and the whole list is two entries - Detail Tweaker LoRA and LowRA 2.0. If you hit "queue" without editing the field, the node will try to look up disney-princess.safetensors, fail with a KeyError: Key disney-princess.safetensors not found in ..., and your run dies. The default is a placeholder example, not a working starting point. Replace it with a real catalog name first.

Install

Same as every node in this pack:

cd ComfyUI/custom_nodes
git clone https://github.com/yolanother/DTAIComfyLoaders

or search "DTAIComfyLoaders" ("Comfy UI Online Loaders") in ComfyUI Manager, then restart. The only dependency is requests.

Where people get burned

• Wrong strength syntax. name:.5:1 works, but watch for stray spaces after commas - "name1:1.0:1.0, name2" is fine, but a space inside a name breaks it.
• First-run downloads. Two LoRAs means two downloads, no progress bars. They're small files, but if a Dropbox link delivers an HTML page instead of the model (it happens with scripted requests), you'll get a load error on a tiny file in models/loras. Fix by grabbing the file manually.
• The catalog is two LoRAs deep. This is a demonstration node more than a production tool. The one genuine advantage over the single-LoRA DTLoraLoader is that it doesn't carry that node's first-run UnboundLocalError bug - this one's the usable half of the pair.

If you want real multi-LoRA muscle with per-slot toggles and trigger-word lookup, rgthree's Power Lora Loader is the community standard for a reason. This node is for the "one field, no extra packs" workflow - just replace the default text before you queue.