YAML LoRA Selector

YAML LoRA Selector is the node that makes the whole pack click if you're doing Wan video. Everything else in ComfyUI-load-image-prompt-lora deals in prompts and name strings; this one emits a WANVIDLORA - the same data type that Kijai's WanVideoWrapper WanVideoLoraSelect node produces - so a YAML-defined LoRA stack can go straight into a Wan 2.1/2.2 workflow with zero hand-placed LoraLoaders per run.

The use case lines up with how the Wan community actually works: people stay on Wan because of its LoRA library, and the character-consistency workflow - a character LoRA trained on a couple dozen stills, swapped in per shot - is the default way to keep a face stable across frames. This node automates the swap: you pick a category, it reads that category's LoRA settings, resolves them against your models/loras, and hands the pipeline a ready-made LoRA stack with weights intact.

How it works

For each of lora1/lora2/lora3 in the category's YAML block, it parses <lora:name:weight> into a name and weight, matches the name against your installed LoRAs (exact, then case-insensitive), and resolves the full file path via folder_paths.get_full_path. Each resolved LoRA becomes a dict shaped exactly like the one WanVideoLoraSelect emits:

{"path": "/.../models/loras/character1_v1.safetensors",
 "strength": 0.8, "name": "character1_v1",
 "blocks": {}, "layer_filter": "", "low_mem_load": False}

blocks and layer_filter aren't supported - you get empty values there, so per-block LoRA targeting isn't possible through this node. The list of these dicts is the wanvid_lora output.

Inputs and outputs

Inputs are category (required - a dropdown from your YAML's keys, or free text if the file wasn't visible at startup) and yaml_path (optional, default setting.yaml). Outputs are wanvid_lora (WANVIDLORA) and yaml_path (STRING, echoed through).

Two ways to wire the output:

• Directly into WanVideoModelLoader's lora input - one YAML category = one LoRA stack applied at load.
• Into WanVideoLoraSelect's prev_lora input - to chain more LoRAs on top of what the YAML defined.

Installing

No dependencies beyond what ComfyUI ships - clone and restart:

cd ComfyUI/custom_nodes
git clone https://github.com/avocadori/ComfyUI-load-image-prompt-lora

Or find ComfyUI-load-image-prompt-lora in ComfyUI Manager.

Common issues

• "No available LoRA files found" warning - your models/loras came back empty from ComfyUI's perspective. Check the files are in the right folder and ComfyUI sees them in a normal LoraLoader.
• The LoRA silently becomes "None". Unmatched names don't crash the run - the node logs a warning and returns a default "None" entry. A category that doesn't exist in the YAML behaves the same way rather than erroring, so check the console if a run looks wrong.
• Strength comes straight from the YAML. If <lora:name:0.3> doesn't look strong enough, that's your number to edit - the selector faithfully passes it through. Remember the community's rule of thumb that a default strength of 1.0 is often too hot; 0.5–0.8 is a common starting range.