Video Reader

If you've ever wanted to run img2img over a clip, do frame-by-frame upscaling, or feed a video back through a workflow to tweak it, the usual ComfyUI answer is "load the frames yourself and pray the colors survive." Video Reader is the part of the ComfyUI-Info-Prompt-Toolkit pack that does it properly: it pulls frames out of a video and drags the generation metadata along, so your video-to-video pass doesn't start from a blank slate.

The name undersells it a little. This isn't just a frame loader - it's the video sibling of the pack's Image Reader, built around the same image_info idea that generation settings should travel with the media so they're reusable later.

How it works

Under the hood it uses PyAV to decode the video into a list of frames, and reads whatever metadata is embedded in the container to reconstruct the generation settings. It handles MP4, WebM, and MKV (the README says MP4; the source happily accepts all three). Frame selection is fully in your control, which matters more than it sounds:

• frame_load_cap - max frames to load. 0 means unlimited.
• skip_first_frames - chop frames off the start (intro cards, empty opening shots).
• select_every_nth - grab one frame per N, i.e. a cheap frame-rate decimation.

The file and directory pickers are Combo dropdowns built from whatever's actually in your input or output folders (path_source chooses which base directory you browse).

Outputs that matter

The node is generous with outputs. The ones you'll actually wire up:

• image - the frames as a list. This is the trap input: it's a list, not a batch. If your downstream node expects a batch tensor, route it through the pack's Image List To Batch first - the README calls this out explicitly.
• audio - the track, if the video has one and your PyAV/FFmpeg build can decode it.
• source_image_info and video_info - the reconstructed generation settings, split into "where the source image came from" vs "how the video was made." These plug straight into the pack's saver nodes so you can round-trip without retyping settings.
• image_infotext(raw) / video_infotext(raw) - the raw metadata strings if you want to inspect or pass them on untouched.

Installing and the PyAV gotcha

Install via ComfyUI Manager (search "ComfyUI-Info-Prompt-Toolkit") or:

cd ComfyUI/custom_nodes
git clone https://github.com/kinorax/comfyui-info-prompt-toolkit.git
cd comfyui-info-prompt-toolkit
pip install -r requirements.txt

Here's the catch that bites everyone: av (PyAV) is not in requirements.txt. The pack deliberately leaves it out so the non-video nodes don't drag in a heavy dependency - the requirement file's own comment says so. Video Reader just fails if it's missing, with the message: "Video reading is unavailable in this ComfyUI environment. Install av first." Fix it once, for both video nodes:

pip install av

Then restart ComfyUI. If you see an error about a codec instead, that's your PyAV/FFmpeg build, not the node - available codecs depend entirely on how your environment was compiled. Also note this pack wants a recent ComfyUI (≥ 0.17.0); on an older install the node won't appear at all.