Lora Remap Strength (QQ)

Some LoRAs just refuse to behave at normal strengths. You slide to 1.0 and nothing happens, or it's only at 0.12–0.2 that the effect shows up at all. That's usually not a broken file - it's an alpha problem. A LoRA's effective contribution is scaled by alpha/rank, and if a LoRA was trained with a weird alpha (or a nonstandard alpha convention, common in some video and distilled-LoRA training), the "100% = 1.0" assumption quietly breaks. LoraRemapStrength is the fix: it rewrites the file so that your target strength becomes the reference.

How it works

You tell it two numbers:

• reference_strength - the strength you've been forced to use to get the effect you want (say 0.15).
• target_strength - the strength you'd rather use (say 1.0).

It computes a scale factor (target / reference), then multiplies every per-module .alpha in the LoRA by that factor and saves a new file next to the original. The math is that simple - LoRA application is alpha-scaled, so scaling alpha shifts the whole effective-strength curve. After the run, applying the LoRA at 1.0 in your loader behaves the way 0.15 used to. The author's tooltips spell out the intent: "Existing effective strength you typically use (e.g., 0.15)" and "Desired equivalent strength (e.g., 1.0)."

The other inputs are secondary. create_missing_alpha (default on) writes an .alpha per module where one is missing, using the rank as the base - worth leaving on, because an alpha-less LoRA is often exactly the one that's been acting weird. output_dtype lets you cast the result to fp16/bf16/fp32 or keep the original. There are no outputs at all: this is an output node that just saves the remapped file, so wire it up, run it once, and use the new file from your loader.

Why you'd actually reach for this

The realistic use case is a character or style LoRA you love but that needs a fiddly non-round strength everywhere, across every workflow you own. Remap it once, and every future workflow treats it like a normal LoRA. It's also quietly useful for harmonizing a stack: if two LoRAs have wildly different alpha conventions, remapping both to 1.0 makes your saved presets read sanely. The metadata comment gets stamped with the remap so you can tell the difference later.

It lives in the pack's KJNodes/lora category - this is the author's adaptation of the KJNodes LoRA utility family, so the flavor is "offline file tool," not "node you leave in your generation graph."

Installing it

Part of siraxe/ComfyUI-WanVideoWrapper_QQ. ComfyUI Manager → search WanVideoWrapper_QQ (or SA-Nodes-QQ), or:

cd ComfyUI/custom_nodes
git clone https://github.com/siraxe/ComfyUI-WanVideoWrapper_QQ.git

Restart ComfyUI, then put the LoRA in ComfyUI/models/loras so it shows in the dropdown.

Gotchas

This node bakes a permanent change into a file - keep a backup of the original unless you're sure. It scales alpha, not the weights, so it's meant for LoRAs that follow the standard alpha-scaling convention; it won't fix a LoRA that's simply undertrained. And the usual pack caveat: the repo was renamed to ComfyUI-SA-Nodes-QQ in v1.3.4, so delete any stale wanwrapper_qq folder if old workflows try to install the wrong thing.