ComfyUI-EagleBridge

ComfyUI-EagleBridge

ComfyUI-EagleBridge adds three output nodes that save generated images, audio, and video from ComfyUI and register them directly in an Eagle library — with in-node previews and generation metadata (prompts, model, seed, …) written into Eagle.

Published on the Comfy Registry as eagle-bridge, so it can be discovered and installed from ComfyUI Manager.

Unofficial community project. This project is not affiliated with or endorsed by Comfy Org or Eagle.

Languages: English · 简体中文 · 한국어 · 日本語

---

What this extension does

ComfyUI-EagleBridge provides three Eagle save nodes, plus the bundled Video Preview utility node:

• Eagle Save Image — saves an IMAGE, sends it to Eagle, can show an image preview, and passes the original IMAGE through its output.
• Eagle Save Audio — saves an AUDIO, sends it to Eagle, and can show an audio player with a seek bar.
• Eagle Save Video — saves a native ComfyUI VIDEO, sends it to Eagle, and can show a video player with a seek bar.

The extension also extracts generation metadata from the upstream workflow — prompts, model, size, seed, steps, CFG, sampler, and more — and writes it into Eagle as tags and annotations. For workflows that compose the prompt only at runtime (e.g. MOSSImageLayoutPrompt), an optional prompt input socket records the exact text.

Requirements

• A recent ComfyUI installation.
• Eagle desktop application running on the same computer.
• An Eagle library must be open while the nodes are executed.
• Eagle's local API must be available at the default address 127.0.0.1:41595.
• Eagle API requests do not use an application-level timeout; a save waits for Eagle to respond.
• Python dependency: requests>=2.31.0.
• Audio/video encoding support depends on the PyAV/FFmpeg/codec support available in your ComfyUI environment.

Installation

ComfyUI Manager: search for ComfyUI-EagleBridge or eagle-bridge and install it directly.

Manual installation:

1. Stop ComfyUI.

2. Copy the ComfyUI-EagleBridge folder into ComfyUI/custom_nodes/.

3. Install the requirements in the same Python environment used by ComfyUI:

   pip install -r ComfyUI/custom_nodes/ComfyUI-EagleBridge/requirements.txt
   

4. Start Eagle and open the library you want to use.

5. Start ComfyUI.

6. If the UI looks stale after updating the extension, hard-refresh the browser (Ctrl+Shift+R) and re-add the node.

Eagle destination folder ID

Every Save node contains an eagle_folder_id text field.

1. In Eagle, right-click the folder you want to save into and choose Copy Link.

2. The copied link will look like this:

   http://localhost:41595/folder?id=MU2LXTTH4KVRF
   

3. Copy only the value after id= — in this example, MU2LXTTH4KVRF — and paste it into eagle_folder_id.

4. Leave eagle_folder_id blank to save to Eagle Library Root.

The folder ID is stored directly in the workflow.

Recording prompts built at runtime (prompt_text)

The Save nodes normally read the prompt text from the workflow graph (CLIPTextEncode and similar text nodes). Some nodes compose the prompt only at runtime — for example MOSSImageLayoutPrompt builds it from layout elements — so no literal text exists in the graph and it cannot be auto-extracted.

For those workflows, every Save node now has an optional input socket named prompt (internally prompt_text) — no text box, just a plug:

1. Connect the STRING output of the node that builds the prompt (e.g. MOSSImageLayoutPrompt → prompt) into prompt_text.
2. The connected text is recorded as the Positive Prompt (or Prompt for audio/video) in the Eagle annotation, overriding auto-extraction.

Leave prompt_text unconnected for normal workflows — nothing changes.

Common filename and folder templates

file_name_template

Default:

%datetime%_%rand%

Supported tokens:

• %date% → YYYYMMDD
• %time% → HHMMSS
• %datetime% → YYYYMMDD_HHMMSS
• %index% → batch index
• %rand% → short random ID

Example filename:

20260915_203128_ab12cd34.webp

local_subfolder

Supported tokens:

• %date% → YYYY-MM-DD
• %time% → HH-MM-SS
• %datetime% → YYYY-MM-DD_HH-MM-SS
• %model% → first detected model name, or UnknownModel if no model can be detected

The folder is always kept inside ComfyUI's output/temp directory.

Eagle Save Image

How to connect it

Connect any ComfyUI IMAGE output directly to the image input. The node also passes the original IMAGE through its right-side output so it can continue into downstream nodes. It remains an output node, so you do not need a separate standard Save Image node just to save to Eagle.

Inputs

• image — ComfyUI IMAGE input. The original value is passed through as the node's IMAGE output.
• eagle_folder_id — Eagle destination folder.
• file_name_template — output filename template.
• local_subfolder — local ComfyUI output subfolder. Default: EagleBridge/Image/%date%/.
• format — webp, png, or jpg.
• quality — 1–100. Used for lossy WebP/JPEG. Default: 95.
• lossless_webp — when enabled, WebP is saved losslessly and quality is not used as the lossy quality setting.
• save_local_copy — when enabled, keep the file in ComfyUI output. When disabled, a temporary file is used so Eagle can import it.
• preview — show the saved image in the ComfyUI node UI.
• memo — optional personal note appended to the Eagle annotation under Memo:.
• website — optional URL stored in Eagle's website field.

Eagle metadata

When available, the annotation is written in this order:

Positive Prompt:
...

Negative Prompt:
...

Model: ...
Size: ...
Seed: ...
Steps: ...
CFG: ...
Sampler: ...
Scheduler: ...

Memo:
...

Unavailable fields are omitted. Eagle tags use the first detected model name only.

Embedded metadata

• PNG: ComfyUI prompt/workflow metadata is embedded when ComfyUI metadata saving is enabled.
• WebP: ComfyUI-style EXIF metadata is embedded when metadata saving is enabled.
• JPEG: workflow metadata is currently not embedded by this extension.

Eagle Save Audio

How to connect it

Connect a ComfyUI AUDIO output directly to audio. The node passes the original AUDIO through its output, so it can remain part of a larger workflow.

Inputs

• audio — ComfyUI AUDIO input.
• eagle_folder_id — Eagle destination folder.
• file_name_template — output filename template.
• local_subfolder — default: EagleBridge/Audio/%date%/.
• format — wav, flac, mp3, or opus.
• bitrate — 64k, 96k, 128k, 192k, 256k, or 320k. Used by MP3/Opus. WAV and FLAC are lossless and ignore this setting.
• save_local_copy — keep the encoded audio in ComfyUI output, or use a temporary file when disabled.
• preview — show the built-in HTML5 audio player with play/pause and seek controls.
• memo — appended to Eagle annotation under Memo:.
• website — optional Eagle website URL.

Prompt and lyrics extraction

For audio/music workflows, the node searches upstream inputs for prompt-like fields including:

caption, prompt, tags, sample_query, description

Lyrics are searched from:

lyrics, lyric

This makes the node useful with ACE-Step-style music workflows and other nodes that use similar field names. If a field cannot be found, it is simply omitted.

Eagle metadata

Typical annotation:

Prompt:
...

Lyrics:
...

Model: ...
Duration: 30 sec
Sample Rate: 48000 Hz
Channels: 2

Memo:
...

Duration is rounded for readability. Eagle tags use the first detected model name.

Notes

• WAV is written as 16-bit PCM.
• Opus may require sample-rate conversion; the extension uses torchaudio when conversion is needed and available.
• Audio files do not currently receive embedded ComfyUI workflow metadata inside the audio container; generation details are stored in Eagle annotations/tags.

Eagle Save Video

How to connect it

Connect a native ComfyUI VIDEO output directly to video. This is intended for current ComfyUI video workflows that expose the native VIDEO socket. The node passes the original VIDEO through its output.

Inputs

• video — native ComfyUI VIDEO input.
• eagle_folder_id — Eagle destination folder.
• file_name_template — output filename template.
• local_subfolder — default: EagleBridge/Video/%date%/.
• format — mp4, webm, or mkv.
• codec — auto, h264, or av1.
• save_local_copy — keep the encoded video in ComfyUI output, or use a temporary file when disabled.
• preview — show an in-node video player with play/pause and seek controls.
• memo — appended to Eagle annotation under Memo:.
• website — optional Eagle website URL.

Codec note

webm + h264 is rejected because that combination is not supported by this node. Actual codec availability also depends on the ComfyUI/PyAV/FFmpeg environment.

Prompt and primary-model extraction

Video workflows often do not use the same text-conditioning path as a standard image KSampler. The node therefore searches common upstream fields such as:

prompt, positive_prompt, positive_text, prompt_text, text_prompt, caption, description, text

It also looks for negative-prompt fields when available.

For the model name, the node tries to select the primary video generation model and demote support components such as VAE, Qwen/CLIP/T5 text encoders, and LoRAs. This is useful for workflows such as MiniMax H3, Wan, and LTX where several model-like files may be present upstream.

Eagle metadata

Typical annotation:

Prompt:
...

Negative Prompt:
...

Model: ...
Size: 960x544
Duration: 5.17 sec
FPS: 24
Seed: ...
Steps: ...
CFG: ...
Sampler: ...
Scheduler: ...

Memo:
...

Unavailable fields are omitted. Duration and FPS are rounded for readability. Eagle tags contain the selected primary model name.

Embedded workflow metadata

The node passes ComfyUI prompt/extra metadata to ComfyUI's native VIDEO.save_to() API when metadata saving is enabled. Whether/how this is embedded depends on the selected container, codec, and ComfyUI implementation.

Troubleshooting

• Player/preview does not appear after updating: restart ComfyUI, hard-refresh the browser (Ctrl+Shift+R), then remove and re-add the node.
• Eagle says a file is duplicated although the filename changed: Eagle can detect identical file content independently of the filename. Generate different content (for example, change the seed) to test.
• MP3/Opus/video encoding fails: codec support depends on the PyAV/FFmpeg stack in the ComfyUI environment.
• Prompt/model is missing: metadata extraction is best-effort. Custom workflows may use field names or graph structures not currently recognized.
• Remote ComfyUI warning: these nodes can cause ComfyUI's backend to write local files and send their paths to Eagle's local API. Be careful when exposing ComfyUI to untrusted remote users.

License

GNU General Public License v3.0. See LICENSE.

---