WanVideo Lora Select From Text (LoraManager)

This is the string-driven version of WanVideo Lora Select - same job, feeding LoRA Manager's library into Kijai's ComfyUI-WanVideoWrapper as a WANVIDLORA, but the LoRA list comes in as a plain multiline STRING instead of the built-in autocomplete box. Plus it adds two things the plain Select doesn't have: LoRA chaining and block-level control.

The string input is the reason to reach for this one. It lets the Wan LoRA selection come from elsewhere in the graph - a Text node, a wildcard, an API field - so you can drive a Wan workflow programmatically or randomize which motion/style LoRAs it uses. If you're building a Wan pipeline you want to control from outside ComfyUI, this is the node that takes the LoRAs as data.

How it works

You put your <lora:name:strength> entries in the multiline lora_syntax box (or wire a string in), and the node assembles a WANVIDLORA for the WanVideoWrapper. The optional prev_lora input lets you chain - pass another WANVIDLORA in and this node's selection stacks onto it. The optional blocks input takes a BLOCKS object so you can apply the LoRA only to specific transformer blocks, which is the advanced knob for controlling where in the model a Wan LoRA acts.

Inputs and outputs that matter

• lora_syntax (STRING, multiline, required) - the LoRA list, usually wired from a text/wildcard source.
• merge_lora (BOOLEAN, default true) - merge weights into the model vs keep separate.
• low_mem_load (BOOLEAN, default false) - lower-VRAM load path.
• prev_lora (WANVIDLORA, optional) - chain onto an existing Wan LoRA selection.
• blocks (BLOCKS, optional) - restrict the LoRA to specific model blocks.

Outputs: lora (WANVIDLORA - into WanVideoWrapper), trigger_words (STRING) and active_loras (STRING).

How to install it

Via ComfyUI Manager: search lora-manager, Install, restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/willmiao/ComfyUI-Lora-Manager.git
cd ComfyUI-Lora-Manager
pip install -r requirements.txt

then restart. As with its sibling, the WANVIDLORA output only plugs into Kijai's ComfyUI-WanVideoWrapper, which you install separately.

Common issues & troubleshooting

The lora output has nowhere to go. Install ComfyUI-WanVideoWrapper - that's the pack that provides the model loader consuming WANVIDLORA, and the blocks/prev_lora types come from that world too.

Chaining didn't stack the way I expected. prev_lora folds this node's selection onto the incoming one; check the order and that both are genuinely WANVIDLORA. For image (non-Wan) workflows you'd use LORA_STACK and the Combiner instead - this chaining is Wan-only.

Block control did nothing visible. The blocks input is a fine-grained tool; if you're not deliberately targeting specific blocks, leave it unconnected. And remember Wan LoRAs transfer only partially across Wan versions, so an odd result may be a version mismatch, not the block setting.