Wan2.2 I2V Lora Loader By Text (LOW)

This is the other half of the Wan 2.2 I2V LoRA story, and for many workflows it's the half that does more of the actual work. Wan 2.2 runs a two-pass denoising scheme: the high-noise model sets motion and composition, the low-noise model refines detail - skin texture, hair, the fine stuff that makes a clip look finished. Wan2.2 I2V Lora Loader By Text (LOW) loads a LOW LoRA, matching it by name from the <lora:...> tags in your prompt text, and applies it to the low-noise model.

If you've read the Wan 2.2 speed-lora debate, this node is the direct tool for the community's favorite compromise: keep the high-noise pass clean and put your speed or style lora only on the low-noise pass. The consensus from people generating "almost 24/7 since release" was that speed loras degrade motion and composition but the low-noise pass tolerates them fine - so the practical setup is a plain HIGH pass and a lora'd LOW pass, which is precisely the split these paired loaders give you.

How it works

Same DNA as the HIGH variant: parse <lora:name:strength> tags from text, fuzzy-match against files in your loras folders (recursive scan via folder_paths.get_folder_paths("loras")), filter by variant and model type, apply. The LOW markers it recognizes are lownoise, lowres, lowfreq, low_noise, low, and ln - same boundary-aware matching, so my_lowres_style.safetensors matches and a random slow in a filename doesn't trip it. model_type_mode (I2V_ONLY default) keeps T2V files out of your I2V pipeline.

The difference from the HIGH loader

One input is missing, and it's the interesting one: the LOW node has no clip_strength_override (only DEBUG_MODE optional). It just applies the lora at its model and clip strengths. If you want per-strength control over the clip leg on the low-noise pass, that's what the full Wan2.2 Lora Loader By Text (the pair-loading node) is for. Here it's a simpler, single-purpose tool.

Outputs: model, clip, lora_info - wire model into the low-noise sampler leg and you're done.

Installing it

ComfyUI Manager → search "illumorae" → install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/CorvaeOboro/ComfyUI_illumorae

Restart. No downloads; it reads your existing models/loras.

Troubleshooting

Same failure signature as the HIGH node: "no match" means the filename lacks a LOW marker or the model-type filter excluded it. Turn on DEBUG_MODE and read the console - it prints the search pattern, how many files it scanned, and the best match score. And one gotcha worth internalizing: if you load a speed lora here but leave the HIGH pass clean, you get the community's favored trade-off; if you also want it on the HIGH pass, don't wire this node there - a LOW lora on the high-noise model is how you get the "flux-level plastic skin" people keep warning about.