AM Read Video

Stock LoadVideo in ComfyUI is a thin thing: mp4 and h264, a limited frame range, no alpha. AM Read Video is the version that reads like a studio I/O node - PyAV under the hood, so it decodes h264, h265, ProRes, DNxHR, VP9 and friends, seeks frame-accurately, handles alpha-bearing codecs like ProRes 4444, and color-manages the result. And its video output socket is the pack's signature trick: wire only that and the decode never even happens.

It's part of comfyui-am-vfx-tools ("AM VFX Tools" category), Adrian Meyer's 13-node VFX toolkit. This is the node to pair with AM Write Video (PyAV on both ends) for a real video round-trip that stock nodes can't do.

How it works

PyAV wraps libav, so you get the whole codec zoo instead of the h264-only special. Frame-accurate seek means a range read seeks to the first frame you want rather than decoding everything up to it. Alpha-bearing pixel formats decode losslessly when the codec supports them - ProRes 4444 and 4444 XQ being the big ones - and the alpha lands on the mask socket per the stock convention (mask = 1 - alpha).

The widgets are the pack's standard I/O set: frame_mode (single/range/all) with before/after edge policies (hold/loop/bounce/black), input_colorspace → working_colorspace via OCIO (with raw_data to bypass), and an optional reformat block. The default input space is Display/Gamma 2.2 Rec.709 - the studio dailies standard - because most video containers don't carry a reliable colorspace tag, so an explicit value is usually more correct than trusting the file.

The trick: the video socket

This is the one worth understanding. The video output is a VideoFromFile that references the source on disk - it doesn't hold pixels. Wire only that socket downstream (say, into an AM Write Image or an API node that takes a file) and PyAV never decodes the frames; peak RAM stays at file-handle level. Wire both image and video in the same graph and you can run a color-managed IMAGE branch for inference and a raw VIDEO branch for an API call that wants the original file - parallel branches, both valid.

Caveat, straight from the author's tooltip: the video passthrough is raw - source colorspace, resolution, codec, untouched by the node's OCIO/Reformat settings. Those apply to image only. If you need the color transform, you either use the image branch or chain an AM transform node on the video.

Outputs

image, mask, audio (or a silent stub), resolved_path, info (dimensions/codec/pixfmt/fps/count), width, height, frame_rate (native, from the PyAV header), frame_count, and video.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/am-pipeline-prod/comfyui-am-vfx-tools.git
cd comfyui-am-vfx-tools
pip install -r requirements.txt

Restart ComfyUI, or search comfyui-am-vfx-tools in ComfyUI Manager. The av>=14.0.0 (PyAV) dependency - and note the alpha support needs PyAV ≥ 11, which the pin covers - comes from that pip line.

Where people get burned

The raw-video surprise: people wire video, run a color transform, and see no change because the video socket is passthrough. That's by design. And if a Linux distro forces a system-FFmpeg link (pip install --no-binary av), verify ffmpeg -h encoder=prores_ks lists the 4444 profiles or alpha encode will silently fall apart. Default decode of a long clip also slurps the whole thing into RAM via the image branch - use the video socket when you only need the file.