LoRA Stack Model Only JK🐉

This is the stack builder to grab when your LoRAs don't touch the prompt side at all. Where the full LoRA Stack JK lets you set a model weight and a clip weight per slot, this one only offers a model weight - the CLIP strength is hardcoded to 0.0. That's not a missing feature, it's the point: video models like Wan and the current generation of text-to-video LoRAs are applied model-only, and forcing clip strength to zero means the stack can't accidentally leak a prompt-side effect into a video generation that has no real text encoder to feed.

How it works

You get six slots, each with three controls:

• lora_1 … lora_6 - a boolean to enable or disable the slot.
• lora_name_1 … lora_name_6 - the dropdown that lists everything in your loras folder.
• model_weight_1 … model_weight_6 - the strength, from -10.0 to 10.0, default 1.0.

The node assembles these into a LORA_STACK - a list of (name, model_weight, 0.0) tuples. A slot only gets added if all three conditions hold: the enable box is checked, the name isn't "None", and the weight isn't zero. Otherwise it's skipped.

There's also an optional lora_stack input: if you feed an existing stack in, the new entries are appended to it. That's how you layer a "base stack" with a per-run stack without rebuilding everything.

Inputs and outputs

Honestly, the only field most people touch is model_weight per slot, and maybe the enable toggles. A quick mental model:

• Check the slot on.
• Pick the LoRA from the dropdown.
• Set the weight. Start at 1.0, drop to 0.5–0.7 if the effect is overpowering.

Output is a single LORA_STACK, which wires into Apply LoRA Stack Model Only JK (or into JakeUpgrade's Wan video subgraphs, which expect model-only stacks by default). Because the tuple carries a clip weight of 0.0, you can also technically feed this into the full Apply LoRA Stack JK - the clip just won't be affected.

Installing it

It ships in ComfyUI-JakeUpgrade:

• ComfyUI Manager → search "JakeUpgrade" → install → restart.
• Or:

cd ComfyUI/custom_nodes
git clone https://github.com/jakechai/ComfyUI-JakeUpgrade
cd ComfyUI-JakeUpgrade
pip install -r requirements.txt

Windows portable users: run install.bat, or ../../../python_embeded/python.exe -s -m pip install -r requirements.txt. No extra model downloads for the node itself - it just reads your existing loras folder.

Common issues

• A slot is enabled but nothing happens. Remember the triple gate: enable box on, a real LoRA selected, and weight ≠ 0. Forgetting to tick the boolean is the classic failure - it's not enough to just pick a name.
• You can't find a clip weight. There isn't one, by design. If a LoRA does need a clip strength for your use case, you want the full LoRA Stack JK + Apply LoRA Stack JK pair, not this one.
• Stack vanishes from a workflow you downloaded. JakeUpgrade deprecated or reshuffled some nodes across v2.x releases. If a JK node is missing after an update, the README points at the deprecated-node settings in the pack's config.ini - worth checking before you assume the workflow is broken.

One more honest note: for a single LoRA this is overkill - a plain LoraLoader is simpler. This node earns its keep in video workflows where you're stacking a style LoRA, a motion LoRA and a character LoRA for the same Wan run, and want to flip one weight without untangling a chain of loader nodes.