LoRA Tuner

If you've ever stacked three LoraLoaders in series and spent ten minutes untangling the wires between them, LoRA Tuner was written for you. It's one box that loads up to eight LoRAs, each with its own strength slider, and hands you back a single MODEL and CLIP to drop straight into your sampler. The Japanese README puts it plainly: it's for people who tune their model's art style with LoRA and don't want the ceremony.

The whole pack is a single ~40-line Python file, MIT-licensed, published on the Comfy Registry, with zero dependencies beyond core ComfyUI. No requirements.txt, no model downloads, no companion megapack. That rarity is the appeal: this is the unopinionated, nothing-to-break end of the multi-LoRA spectrum.

How it works

The mechanism is almost comically honest. The node constructs eight instances of ComfyUI's built-in LoraLoader and chains them in series. For each slot 0→7, if a LoRA is selected and its strength isn't 0, it pipes model+clip through load_lora(model, clip, name, strength, strength). Two real consequences follow:

• Order matters. Slots apply in sequence, so the last active LoRA touches the weights last and has the final say. Blending a character LoRA with a style LoRA? Think about which one you want dominant and put it later in the chain.
• One strength drives both MODEL and CLIP. There's no separate clip strength here, the way some loaders split it. If a LoRA is fighting your prompt, your only dials are the single slider or skipping the slot.

The inputs that matter

Feed it a MODEL and a CLIP (from your checkpoint loader), then for each of the eight slots:

• N:lora - a dropdown of every LoRA file in your loras folder. Leave it blank to skip the slot.
• N: - a strength slider, -2.0 to 2.0, default 1.0. Set it to 0 to mute a slot without clearing your pick. Negative strengths invert a LoRA's effect - handy when something overshoots and you want to push back instead of deleting it.

Outputs are a MODEL and a CLIP. Wire them into your sampler's model, positive and negative inputs exactly as you would from a plain LoraLoader.

Install

Easiest via ComfyUI Manager - search the pack title "ComfyUI-LoRA-Tuner" and hit install. Or the manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/AonekoSS/ComfyUI-LoRA-Tuner

Restart ComfyUI and you're done. No pip step, no weights to fetch; the node appears under utils as "LoRA Tuner".

Honest caveats

This is the minimalist option, so know the tradeoffs. There's no trigger-word display, no CivitAI integration, no reordering (slots are fixed at eight, always applied in order), and no per-slot bypass except setting strength to 0. If you want the full-featured alternative, rgthree-comfy's Power Lora Loader is the community standard - it reads trigger words off files, fetches them from CivitAI, and generally lives up to its reputation for big stacks. The output is equivalent, because both ultimately call the same core LoraLoader math.

So the pragmatic take: if you want two or three LoRAs in one node without installing a megapack, this is it. If you already run rgthree, you don't need it. And if you wanted a fancy LoRA manager, that's not what this is - it's a loader, and a clean one.

One last thing: the pack's only JS is a cosmetic tweak that gives the slider widgets breathing room, so it tends to stay out of the way when ComfyUI's own frontend updates roll through - less breakage surface than the custom-widget behemoths, which is a quiet blessing.