Zenkai Scene Video V1

If you've ever stared at Video Helper Suite's Load Video and wished it understood "grab me the third clip from that folder," this is the node for you. Zenkai Scene Video V1 is DJZ-Nodes' folder-first video loader: instead of typing a path, you pick a folder and let the node decide which clip you get - by index, by rotation through the folder, or by seed.

It sits in the same family as Zenaki Video Prompt V1 (note: different spelling, both exist in the pack). This one is the version without the prompt-matching logic - no default_prompt input, no prompt output. If you need the prompt to ride along with the clip, reach for the Zenaki variant instead. If you just want frames, audio, and file info from an organized scene folder, this is the leaner choice.

How it works

You keep your source clips in folders under custom_nodes/DJZ-Nodes/scenevideo/ - one folder per scene, a few clips inside. The dropdown populates itself from those subfolders. The mode input decides which clip gets loaded:

• single_video - pick exactly one clip with index
• incremental_video - step through clips as you bump index, wrapping around
• random - use seed to land on a clip

The decoding path is the same as the rest of DJZ's video tooling: OpenCV reads frames as float tensors, ffmpeg extracts audio (only if something actually consumes it), and memory_limit_mb auto-computes how many frames you can safely hold in RAM using psutil. That auto-memory behavior is the practical trick here - it means you can feed a long clip without OOM-killing your sampler, because the node loads what fits, not the whole file.

The inputs and outputs you'll actually set

• folder - dropdown of scenevideo/ subfolders; shows none until you add one
• mode - single_video, incremental_video, or random
• index / seed - the selector, depending on mode
• skip_frames / max_frames - cut the intro and cap the load (0 = all)
• force_rate - force a frame rate instead of trusting the file's metadata
• meta_batch - optional VHS_BatchManager for frame-by-frame batching with Video Helper Suite

Outputs: frames (IMAGE), frame_count (INT), audio (AUDIO), video_info (VHS_VIDEOINFO), and filename_text (STRING). That's the full return set - frames into a VAE encode or an img2vid model, video_info into VHS nodes for timing, filename_text for naming your renders after the source clip. Like its sibling, the vae input is declared but unused in the current code; frames come out decoded.

Install

ComfyUI Manager → search DJZ-Nodes, or:

cd ComfyUI/custom_nodes
git clone https://github.com/MushroomFleet/DJZ-Nodes
cd DJZ-Nodes
pip install -r requirements.txt

Restart ComfyUI after. You also need Video Helper Suite installed for the VHS_BatchManager and VHS_VIDEOINFO types to resolve, and ffmpeg on your system for audio. Two DJZ-specific traps: psutil is imported but not in the pack's requirements.txt, and the pack's __init__.py silently skips nodes whose imports fail. If the node isn't in your list, open the console - "Unable to import ZenkaiSceneVideoV1" means you're missing a dependency (pip install psutil fixes the common case).

Troubleshooting

The scenevideo/ folder is created on first run, so an empty dropdown is normal until you add a subfolder with videos inside. "No video files found" usually means you dropped files straight into scenevideo/ instead of a subfolder, or your extension isn't one of .mp4/.avi/.mov/.mkv. And if a fresh clone hits a GitHub 404, that's a known transient blip from mid-2025 - retry the clone rather than debugging your setup.