EA Power LoRA +CLIP

The model-only EA Power LoRA is great for WAN 2.2, but for SD/SDXL/Flux pipelines the LoRA needs to hit the text encoder too - otherwise your character and style LoRAs half-work because their CLIP weights are zero. EA Power LoRA +CLIP is the sibling that adds a CLIP path: one row-based panel, two weights per row (model and CLIP), and both a model and a clip output.

It's the "full pipeline" version of the pack's LoRA stacker, and it covers the classic image-generation case the model-only node deliberately skips.

How it works

Mechanically it's the model-only node plus a text-encoder channel. The model input is required; clip is optional - you can feed it a CLIP from a Checkpoint Loader or CLIP Loader. Rows carry strength_model and strength_clip, and the web UI gives you a per-row checkbox plus both weight fields. At execution it chains ComfyUI's core LoraLoader per row, passing the CLIP through only when one is wired in.

The behavior worth knowing: set a row's CLIP weight to 0.00 and that LoRA becomes model-only. This is the escape hatch the README calls out explicitly - you can run a mixed stack where a style LoRA is model-only and a character LoRA gets real CLIP weight, all in one panel. No need to split into two nodes.

Same quiet behaviors as the model-only version: missing LoRA files are silently ignored, and row order doesn't change the result (application is additive).

The inputs and outputs that matter

• model (required) - your checkpoint's model.
• clip (optional) - the text encoder. Wire it or don't; the node behaves either way.
• loras_json - the hidden serialized rows, normally manipulated through the panel UI.

Outputs are model and clip, both pass-through ready for your sampler and CLIP Text Encode. Per row, the two weights are the thing you'll fiddle with: model strength first, CLIP strength second. The README's practical guidance: many style LoRAs work fine with CLIP low/off, while character/concept LoRAs often benefit from CLIP weight - so the per-row split is where this node earns its keep.

When to pick this over the model-only node

• SD / SDXL / Flux: use this one - those pipelines have a real text-encoder path.
• WAN 2.2 UNet-only branches: use plain EA Power LoRA; there's no CLIP socket to feed here.
• Mixed stacks: this node handles them, via the 0.00 CLIP-weight trick.

Install

Part of comfyui-ea-nodes: ComfyUI Manager → search comfyui-ea-nodes, or git clone https://github.com/ExoticArts/comfyui-ea-nodes into custom_nodes/, restart, hard-refresh. No extra Python dependencies; LoRAs load from ComfyUI/models/loras.