Studio42 Video PatchLift Loader

Studio42 Video PatchLift Loader is the video version of the pack's PatchLift Loader: it loads a video, turns it into frames, and extracts the same patch region across every frame - with a per-frame visual overlay so you can see exactly what region is being tracked. Pair it with the pack's Video PatchDrop and you can lift a region from a clip, process it, and drop it back with temporal consistency.

This is the node you reach for when the thing you want to edit doesn't stay still - a face that moves across the frame, a logo, a region that needs per-frame effects. It's not optical-flow tracking; it's a fixed region applied consistently across frames, which is the honest, predictable version of the job.

What you set

• video - the file to load (MP4, WebM, MKV, AVI, MOV, GIF per the pack's docs).
• patch_x / patch_y - the region's position in the frame, 0–4096.
• patch_width / patch_height - 16–1024.
• patch_shape - rectangle, circle, rounded_rectangle, with corner_radius.

The video-control extras matter for long or heavy files:

• frame_load_cap - limit total frames loaded (0 = unlimited). The memory saver.
• skip_first_frames - drop the first N frames.
• select_every_nth - sample every Nth frame (e.g., 2 = half the framerate).
• force_rate - override the frame rate.
• custom_width / custom_height - resize the frames on load.

And the selection visuals, mirroring the image version: grid_size, grid_opacity, selection_color, selection_thickness, show_coordinates, corner_radius, preview scaling.

What comes out

Ten outputs - the useful ones: video_frames (all frames as an IMAGE batch), preview_frames (with the overlay), patch_frames (the extracted patch per frame), patch_masks (per-frame masks for non-rectangular shapes), the patch_x/y/width/height INTs, plus fps, frame_count, and audio (the video's audio track, if any, as AUDIO). That last one is a nice bonus - you don't need a separate audio loader for a video file.

The honest limits

The patch region is fixed across frames, so it works great when the region stays put and drifts when it doesn't. For tracking a moving subject, you'd want a tracker feeding per-frame coordinates instead. Within its scope - "same region, every frame" - it's exactly right, and frame_load_cap makes it usable on long clips.

Installing it

Same pack, 24oiduts-ComfyUI:

• ComfyUI Manager: search 24oiduts, install, restart.
• Manual:

  cd ComfyUI/custom_nodes/
  git clone https://github.com/GeekyGhost/24oiduts-ComfyUI
  pip install -r 24oiduts-ComfyUI/requirements.txt

  then restart.

This is the node with the heaviest real dependency list: it needs opencv-python and ffmpeg-python, plus a system FFmpeg for the codec work. The pack's __init__.py even tells you when it can't load - the whole video section disappears with a "install opencv-python, ffmpeg-python" message. The GitHub README is a stale Studio42 template; don't let its WIP banner scare you.

Troubleshooting

• Video nodes missing from the menu - the pack's video section failed to import; install opencv-python and ffmpeg-python, plus system FFmpeg, then restart.
• Out of memory on long clips - set frame_load_cap and select_every_nth.
• Patch drifts from the subject - expected with a fixed region; keep the region locked to a stationary part of the frame or add per-frame coordinates upstream.

Lift a region across a whole clip, keep the audio, and hand it to Video PatchDrop - that's the pipeline this node exists for.