Load Video

DeepGen_LVID is the pack's video loader, and like DeepGen_LIMG it exists to do two things: hand a video into the graph as a proper VIDEO type, and pull the XMP metadata out of the file - description, dialogues, and assets - the way the image loader does. For anything where a previous DeepGen generation's context needs to come back into the workflow, this is the door.

Why you'd reach for it

Two jobs, honestly. First, feeding video back into the pack: the element-based video node (DeepGen_I2VR) takes a reference video input, and this loader is the natural source for it - load a clip, wire it into an element slot, and the model has a movement reference. Second, round-tripping metadata: load a generated video, grab its description, and feed it into a new generator. Beyond that, it also gives you a force_height rescale that the graph can't do natively on a video file. If you just need to play a video in ComfyUI with no pack involvement, the ecosystem's other video utilities cover that - this one is tuned for DeepGen's own data.

How it works

The video is picked from ComfyUI's input/ directory (mp4, mov, webm, mkv, avi, m4v). If force_height is set and differs from the file's height, the node resizes it with OpenCV (keeping aspect ratio, ensuring even width) and caches the result in ComfyUI's temp dir. The video is then returned as a ComfyVideoMock - the pack's stand-in for ComfyUI's VIDEO type - and the metadata fields are read from the file's XMP (or native sidecar data). Pure local file work; no API key, no credits.

The inputs and outputs

• video - dropdown of video files in input/, with upload.
• force_height (0–8192, step 8, default 0) - target height; width scales to match. 0 = original size.
• video (VIDEO) - the loaded clip, as a ComfyVideoMock that the standard video preview/save nodes understand.
• description, dialogues, assets (STRING) - the metadata, empty strings when absent.

Install

Same pack, same steps:

cd ComfyUI/custom_nodes
git clone https://github.com/deepiksdev/ComfyUI-DeepGen-API.git
cd ComfyUI-DeepGen-API
pip install -r requirements.txt

Restart ComfyUI. No API key needed for loading.

Gotchas

This is where the pack's opencv-python dependency earns its keep - if the resize path errors (an odd codec, a corrupt file), the node falls back to the original file rather than failing, which is nice, but it also means a silent fallback you might not notice. Metadata only exists if the file was written with it; third-party clips return empty strings, so don't build a workflow that assumes assets is always populated. And if a downstream node doesn't recognize the video output, remember it's a mock object wrapping a temp file path - the pack's own video nodes accept it, and standard video save nodes handle it, but an exotic third-party video node may not.