comfyui-ea-nodes (ExoticArts)

Custom nodes for ComfyUI focused on quality-of-life workflows, WAN/Flux-friendly LoRA stacking, and simple IO utilities.

Current nodes

• EA Power LoRA (EA_PowerLora) — Apply N LoRAs (in order) to a MODEL and optional CLIP.
• Trim Images (Start/End + Previews) (EA_TrimImagesStartEnd) — Trim image sequences and expose first/last-frame previews and count.
• EA Filename → Combine (EA_FilenameCombine) — Build safe filename prefixes and absolute path stubs for Save nodes.

---

Installation

Comfy Registry (recommended)

1. In ComfyUI-Manager, open Custom Nodes → Registry.
2. Search for EA Comfy Nodes (Publisher: ExoticArts) and install.
3. Restart ComfyUI.

If you see “Install Missing Nodes” complaining about EA_PowerLora, update ComfyUI-Manager to the latest and/or install via URL as below.

Manual / via URL

• Manager → Install via URL: paste the repo URL.

• Git:

  cd ComfyUI/custom_nodes
  git clone https://github.com/ExoticArts/comfyui-ea-nodes
  

  Restart ComfyUI.

---

Nodes

EA Power LoRA

ID: EA_PowerLora • Category: EA / LoRA

Apply multiple LoRAs, in order, onto a MODEL (and optionally a CLIP). The node ships with a small web UI that lets you add/remove rows and set strengths per LoRA.

Inputs

• model (MODEL) — Required base model.

• clip (CLIP) — Optional. If present, LoRA clip deltas will be applied.

• loras_json (STRING) — Hidden; the web UI maintains this. Format:

  [
    {"name": "example.safetensors", "strength_model": 0.8, "strength_clip": 0.8},
    {"name": "hands_fix.safetensors", "strength_model": 0.6, "strength_clip": 0.0}
  ]
  

Outputs

• model (MODEL) — Model with all LoRAs applied.
• clip (CLIP) — Updated CLIP (or None if no CLIP was provided).

Behavior & notes

• LoRAs are resolved against the ComfyUI /models/loras list; unknown names are skipped.
• If your workflow has no CLIP (e.g., WAN 2.2), only strength_model applies; strength_clip is ignored.
• The node re-executes whenever loras_json changes.

Quick usage

1. Drop EA Power LoRA after your CheckpointLoaderSimple.
2. Click ＋ Add LoRA, select a filename from the dropdown, set strengths.
3. Continue into your sampler/vae/decode stack.

Minimal WAN template A ready-to-load blank graph is included:

• examples/wan/ea_wan_power_lora_blank.json

It wires: CheckpointLoaderSimple → EA_PowerLora and waits for the rest of your pipeline.

---

Trim Images (Start/End + Previews)

ID: EA_TrimImagesStartEnd • Category: EA / Video

Trim leading/trailing frames from an image batch [N, H, W, C] and expose first frame, last frame, and frame count for convenient previews and diagnostics.

Inputs

• images (IMAGE) — A batch of images (e.g., from Load Images, VAE decode, or a video frames node).
• skip_first (INT) — Number of frames to remove from the start (default 0).
• skip_last (INT) — Number of frames to remove from the end (default 0).

Outputs

• images (IMAGE) — Trimmed sequence.
• first_frame (IMAGE) — First frame after trimming (shape [1, H, W, C]).
• last_frame (IMAGE) — Last frame after trimming (shape [1, H, W, C]).
• frame_count (INT) — Number of frames remaining.

Example

• LoadImages → EA_TrimImagesStartEnd(skip_first=2, skip_last=2) → SaveAnimatedImage
• Use first_frame/last_frame to quickly preview edges with Preview Image or a separate save.

---

EA Filename → Combine

ID: EA_FilenameCombine • Category: EA / IO

Builds a safe filename prefix (POSIX-style) plus an absolute path stub you can feed to save nodes or other IO steps. Helpful when you want reproducible naming for multi-step pipelines.

Inputs

• subfolder (STRING) — e.g., trim/6_belle. Will be normalized (no ..).
• stem (STRING) — Base name, e.g., my_video.
• suffix (STRING) — Extra piece appended after stem, e.g., _trim.
• use_video_info (BOOLEAN) — Whether to append video_info.
• video_info (STRING, optional) — Extra token (e.g., fps24_1920x1080).

Outputs

• prefix_for_combine (STRING) — For SaveImage.filename_prefix (e.g., trim/6_belle/my_video_trim_fps24).
• fullpath_stub (STRING) — Absolute path without extension, useful for scripted saver nodes.

Example

• EA_FilenameCombine → SaveImage(filename_prefix)
• EA_FilenameCombine → SaveAnimatedImage(filename) (append extension in that node as needed)

---

Example Workflows

1) WAN + Power LoRA (blank starter)

• examples/wan/ea_wan_power_lora_blank.json

• Steps to complete:

  1. Install/point to your WAN checkpoint in CheckpointLoaderSimple.
  2. Add your usual conditioning/sampler/vae/decode chain.
  3. In EA Power LoRA, add rows and strengths.

2) Trim a sequence and save

LoadImages → EA_TrimImagesStartEnd(skip_first=8, skip_last=8)
           ↘ first_frame → SaveImage(prefix="debug/edges/first")
             last_frame  → SaveImage(prefix="debug/edges/last")
           → SaveAnimatedImage(filename="output/clip_trimmed.mp4")

3) Build a stable filename prefix

EA_FilenameCombine(subfolder="runs/0420", stem="my_clip", suffix="_trim",
                   use_video_info=true, video_info="fps24_1920x1080")
   → SaveAnimatedImage(filename_prefix)

---

Developer notes

• The nodes are designed to be import-safe in CI (no torch/Comfy required at import time). Heavy deps are lazy-imported inside the node methods.

• A simple validator lives in tests/validate_ea_nodes.py.

  • Locally: python tests/validate_ea_nodes.py
  • CI: GitHub Action runs on each push.

• Web component for EA Power LoRA lives at web/ea_power_lora.js. It fetches LoRA filenames from /object_info and stores all rows in the hidden loras_json widget.

---

Troubleshooting

• “Install Missing Nodes can’t find EA_PowerLora”

  • Make sure the pack is published in the Comfy Registry, or install via URL.

• WAN without CLIP

  • The node works fine; only strength_model applies.

• LoRA not in dropdown

  • Ensure the file is in ComfyUI/models/loras and restart ComfyUI.

---

License

MIT — see LICENSE.