RedNode LoRA Stack

Running three LoRAs in ComfyUI means three LoraLoader nodes, three places to type strengths, and a prompt where you've hand-copied each trigger word from its model card. RedNode LoRA Stack is the version that bundles the whole thing: multiple LoRA slots on one node, per-slot strength, optional random strength ranges, trigger-word output for prompt building, and named presets for the entire stack. It's the most-used kind of utility in this pack - model loader that replaces a wall of boilerplate.

The inputs are the story. A model goes in, and you get model and clip back out (wire the optional clip input to get CLIP-side strengths applied too). The stack_json box holds the slot list - the panel above the node edits it, but it's plain JSON, so you can hand-edit or copy it between workflows. The seed drives randomized slots: same seed plus same stack equals the same strengths, and if you set control_after_generate to randomize, each queue rolls a fresh combination. That's a genuinely nice workflow - load a stack, let it vary within ranges, hunt for a good roll.

The two outputs that aren't on a stock loader are what make it worth it. keywords hands you the trigger words as a string, ready to splice into your prompt - no more copying them from model cards. applied_loras reports what actually got applied and at what strengths, which doubles as your console record of a run. It's the "row-list LoRA UI" idea from no8d's ComfyUI-NO8D-controls, which the README credits as inspiration, implemented inside a single node with the stack concept bolted on.

Where people actually get value vs. where it's just nice: if you have one or two fixed LoRAs, a stock loader is fine and this is overhead. The stack pays for itself the moment your workflow has a rotating cast - character LoRA plus style LoRA plus a vae tweak, with the strengths changing by experiment. Random ranges are the feature you didn't know you wanted until you stop hand-tweaking three numbers per run.

Two gotchas to carry in. First, presets: the preset dropdown loads a saved stack, and newly saved ones appear after a node-definition refresh (r) - the pack's standing refresh contract. Second, randomized strengths need the seed to actually change between runs, which means understanding control_after_generate (fixed/randomize/increment) on that seed widget - set it to randomize if you want a fresh roll each queue, and remember that ComfyUI's default is to change the seed after the run, so the value you see is the one that will be used next, not the one that just ran.

Installation is the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/RedNodeAI/ComfyUI-RedNodeStudio.git

or search RedNode Studio in ComfyUI Manager, restart. No pip dependencies, works with any model.