CI

ComfyUI EA Nodes

Compact, user-friendly LoRA stacking nodes for ComfyUI. Designed to work cleanly with WAN 2.2 (no CLIP path) and with SD/SDXL (model + CLIP).


Nodes

EA Power LoRA

Apply one or more LoRAs to the MODEL only (no CLIP path).

Use this when: your pipeline has no text-encoder/CLIP path (e.g., WAN 2.2).

Wiring

[Model Loader] → (model) → [EA Power LoRA] → (model) → […]

EA Power LoRA +CLIP

Apply one or more LoRAs to MODEL and, optionally, CLIP. Includes a global “Apply to CLIP” toggle at the top.

Use this when: your pipeline includes a text encoder (e.g., SD/SDXL).

Wiring

[Checkpoint Loader] → (model, clip) → [EA Power LoRA +CLIP] → (model, clip) → […]

Apply to CLIP

  • On: each row’s CLIP strength is used.
  • Off: CLIP strengths are ignored (model-only behavior), but the CLIP socket stays wired for convenience.

UI at a Glance

  • + Add LoRA: add a row.
  • On: enable/disable a row.
  • LoRA: dropdown of files in /models/loras (falls back to text if the list can’t be fetched).
  • Model: UNet/model strength (0.0–2.0).
  • CLIP (only in +CLIP): text-encoder strength (hidden when “Apply to CLIP” is off).
  • ✕ Remove: delete the row.

Rows are top-aligned with a small gap between entries for readability.


Notes & Tips

  • Which node to choose:

    • WAN 2.2: use EA Power LoRA (no CLIP path to patch).
    • SD/SDXL: use EA Power LoRA +CLIP and tune the global toggle + per-row strengths.
  • LoRA order: runtime LoRA application is a linear sum of deltas; order does not change the result.

  • Trigger words: they act through the text encoder; some LoRAs (character/concept) benefit from CLIP strength, while many style LoRAs work fine with CLIP low or off.

  • Missing files: unknown LoRA names are safely ignored.


Install

Comfy Manager / Registry: search comfyui-ea-nodes and install. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/ExoticArts/comfyui-ea-nodes
# Restart ComfyUI and hard refresh the browser (Ctrl/Cmd+Shift+R)

License

See LICENSE.

S
Description
No description provided
Readme MIT
1.5 MiB
Languages
Python 74.8%
JavaScript 25.2%