EagleFeeder (AnimatedWEBP)

Animated WebP is the quietly superior middle child of the animation formats: smaller than a GIF, no player needed like an MP4, and Eagle handles it natively. This node is SaveAnimatedWEBP with a destination change - it takes your sequence of frames, encodes them into an animated WebP, and feeds the file straight into a running Eagle app with tags and a folder of your choosing. It's the second of three nodes in the ComfyUI Eagle Feeder pack (PNG, animated WebP, MP4), and if your workflow already ends in a looping preview, this is the one you'll reach for.

It replaces the manual save-and-drag dance completely. Frame batch goes in, animated WebP shows up in Eagle, tagged, filed. Same inputs as the stock node plus the Eagle plumbing - which is exactly what the README means when it says everything else "behaves the same as ComfyUI's standard SaveAnimatedWEBP."

How it works

Mechanically it's the pack's one trick, applied to a new format: the first time any Eagle Feeder node loads, the pack starts a tiny local file server on port 8000 serving a scratch folder inside custom_nodes/comfyui-eagle-feeder/images/. On run, the node turns each frame tensor into a PIL image, saves them all into a single animated WebP, then calls Eagle's /api/item/addFromURL with a http://localhost:8000/<filename> URL plus your tags and folder. Eagle fetches the file and imports it. Old scratch files are purged after two days, so it doesn't silently accumulate. For a remote Eagle (different machine, container), the file server binds 0.0.0.0 but the URL it hands Eagle is always localhost:8000 - bridge that port with an SSH tunnel and you're set; on the same machine it just works.

The inputs that matter

• images - the IMAGE tensor where each entry is a frame, straight off your AnimateDiff-style output or whatever produces the sequence. Order in the batch is order in the animation.
• fps - default 6, which is the classic low-fps animation sweet spot. Range is a generous 0.01 to 1000.
• lossless - on by default. Leave it on for clean loops; flip it off and quality (default 80) actually starts mattering for file size. The classic confusion: people dial quality down, see no change, and it's because lossless is still on.
• method - default / fastest / slowest. This maps straight onto Pillow's WebP encoder methods, so slowest compresses better at the cost of encode time. For big batches, fastest is a legitimate choice.
• tags, folder_name, eagle_host, eagle_token, embed_workflow - the shared Eagle plumbing from this pack. tags is comma-separated and forceInput, so wire in your prompt or a WD14-Tagger output. folder_name must already exist in Eagle (no auto-create, empty = root, duplicate names = arbitrary pick). The token comes from Eagle's Preferences → Developer. embed_workflow (on by default) tucks the prompt into the WebP's EXIF so the loop is still reconstructable later.

Install

ComfyUI Manager, search "ComfyUI Eagle Feeder," or:

cd ComfyUI/custom_nodes
git clone https://github.com/tighug/comfyui-eagle-feeder.git

Restart. No model files, no big downloads. One wrinkle: requirements.txt pins torch==2.6.0 exactly, which can fight a ComfyUI install that already has a different torch build - skip the dependency step if Manager starts proposing a torch change; the node runs fine on whatever ComfyUI already loaded.

Where it trips people

Eagle has to actually be running, or the folder lookup dies on the first call. A missing token (Preferences → Developer) or an occupied port 8000 (no fallback - the file server fails silently and uploads go dead) are the other two usual suspects. And if you were expecting GIF-like size savings, remember: quality only does anything with lossless off. Otherwise it's a boring, reliable end-of-pipeline node - which is exactly what you want an output node to be.