Camera Events To Images

See What Your Event Stream Actually Caught

If EVTVideoToEvents is the pack's camera, this node is the monitor you point at it to see whether anything useful is happening. EVTEventsToImage takes an EVT_EVENTS stream and renders it back into a viewable image - red where brightness went up (positive polarity), blue where it went down, green left alone. It's a sparse plot of where and when things changed, not a photograph, and it's not pretending otherwise.

Why you'd reach for it

The upscaler can silently produce mush, and when it does, the first suspect is the event data. This node lets you actually look at it. The pack's own example workflow keeps one in the graph as a live side-channel: events get rendered to images and saved alongside the upscaled video, so you can watch what the "camera" caught in parallel with what the model did with it. If your event frames look nearly empty, you've found your problem before the upscaler wastes ten minutes confirming it.

The inputs and output

• events (EVT_EVENTS) - the stream from EVTVideoToEvents.
• width (INT) and height (INT) - your frame resolution. These have to match what the events were generated from, or events plot in the wrong place. The example workflow wires GetImageSizeAndCount into both so they can never drift; you should too, unless you like chasing phantom bugs.

Output: images (IMAGE) - one visualization per event batch, which you can send to any video-save node (the example uses VHS_VideoCombine) or just eyeball in the UI.

Install

Same pack, same install: ComfyUI Manager → search "ComfyUI-EvTexture", or clone the repo into custom_nodes and restart.

cd ComfyUI/custom_nodes
git clone https://github.com/tocubed/ComfyUI-EvTexture

The pack's dependencies are just torch/torchvision, already present. Nothing to download for this node.

Common issues

This is a debug node, so "issues" mostly means "your events are boring." Dark or slow footage gives you a mostly black frame - that's correct behavior, and it's diagnostic. Keep in mind it sits off the actual inference path: the upscaler never touches this node's output, so you can delete it from the graph once you trust your events. And don't mistake the red/blue plot for an edge map you can feed anywhere else - the IMAGE type is generic, but what's inside is an event visualization, not a frame that makes sense to a normal video model.