ITDA Load Export

ITDA is a whole video editor living inside ComfyUI - a Korean word (잇다) meaning "stitch," which tells you exactly what it's for. AI video means generating several clips and joining them, and the join is the fiddly part. ITDA exists to do that joining in a real timeline with layers, audio, and auto-cut suggestions. ITDA Load Export is the closing half of that loop: after you click Export in the editor, this node reads whatever file landed on disk and hands it back to you as actual ComfyUI data - frames as an IMAGE batch, plus the mixed audio.

The design choice worth appreciating: it's decoupled from the editor by design. No live connection, no websocket handshake. Each export writes a small last_export.json manifest into ComfyUI/output/ITDA/<project>/, and this node just reads that. Run the queue whenever you want and you always get the most recent export - you don't have to babysit state. That's the same "read the file on disk" pattern as half the loaders in the ecosystem, and it's exactly right for a node whose job is to be boring and reliable.

The inputs that matter

Only two, and you'll usually touch just one.

• project_name - a dropdown of your ITDA projects, read live from ComfyUI/input/ITDA/projects. Pick the project whose latest export you want.
• project_name_override (optional) - a plain STRING input that takes priority over the dropdown when connected. It exists for a plumbing reason: ComfyUI combo inputs can't accept a direct STRING wire in current builds, so this is where you connect ITDA Open Editor's project_name output. Wire it up and the dropdown becomes a fallback you can ignore.

What comes out

Four outputs, and you'll reach for two of them constantly.

• frames (IMAGE) - every frame of the export as one batch. This is the thing you feed an upscaler, a video model's first-frame, a preview node, or a save-to-video path.
• audio (AUDIO) - the mixed track, if the export has one. Wire it to any audio-capable output.
• frame_count (INT) and fps (FLOAT) - metadata, useful when you need to tell a downstream node how long the video is or what frame rate it was cut at. The fps is the timeline's export rate, which is handy because AI video frame rates are whatever your model happened to produce.

Installing it

The whole pack comes as one install - there's no separate package for this node.

cd ComfyUI/custom_nodes
git clone https://github.com/designloves2/itda
# restart ComfyUI

Or skip the terminal: ComfyUI Manager → search ITDA → install. That's the easy route, and it's the same one you use for everything else in this ecosystem.

Good news on dependencies: it's light. aiohttp and numpy (both already in ComfyUI's environment) are the only hard requirements. ffmpeg and ffprobe must be on your PATH - standard ComfyUI installs ship them, but a trimmed Linux setup is where people hit this. OpenCV (motion matching) and librosa (beat detection) are optional and degrade gracefully if missing.

Where people get burned

The most common failure is also the most obvious: the "No export found" error means you haven't clicked Export in the editor for that project yet. The node validates this before the queue even starts (a nice touch - it fails fast with a readable message instead of erroring mid-run), but the message only tells you what's missing, not what to do: open the project in ITDA, click Export, then run the queue again.

Two smaller gotchas. If you've wired project_name_override, the dropdown is silently ignored - set the override right or you'll chase a wrong project. And because this node reads "whatever the last Export produced," re-exporting while you have a queue stacked up will change what the next run picks up. That's a feature, not a bug, but it's worth knowing before you get confused why a saved workflow suddenly outputs different footage. For a node whose entire job is one honest handoff, it does exactly what it says - no API keys, no remote service, just a file read from your own disk.