◎ Radiance Read

ComfyUI's built-in VHS or Load Image will get you video frames or a single image, but if you work with footage - real clips, EXR plates, image sequences, log footage - you end up stitching together three different loaders and remembering which one handles what. Radiance Read is the unified front door for that: one node that reads video files, image sequences, and single images (including 32-bit EXR with depth and alpha channels) and hands you the frames, a mask, frame count, dimensions, fps, and even audio.

It's the read half of the pack's "Universal Digital Cinema I/O" story, and it's the piece that makes the rest of the suite make sense. Everything downstream - the grade nodes, the scopes, the false-color exposure check - assumes you've got clean frames with correct metadata, and this is how they arrive.

What you're setting

• source_path - a file path or sequence pattern. Videos: mp4, mov, gif, webp, avi, mkv, webm. Plus EXR (with Z/depth and A/alpha channels read out), HDR, and image sequences.
• start_frame - where to begin (1-based). Skipping the first N frames of a long clip is the obvious use.
• frame_limit - how many frames to load; 0 = load all. This is the RAM control. A 4K video at 24fps is heavy, and loading a thousand frames into VRAM is how you OOM. Set a limit for look development, expand for the real pass.
• input_colorspace - the one to actually think about. Radiance is a color-science pack, so it wants to know what space the footage is in: sRGB, Linear, ARRI LogC3/LogC4, Sony S-Log3, Panasonic V-Log, DaVinci Intermediate, ACEScg, ACEScct. If your clip is log footage, declaring it here means the downstream grade and HDR nodes interpret it correctly instead of treating it as gamma-encoded sRGB and mangling your contrast.
• fps_override - 0 = auto-detect from the video (or 24 for a sequence). Set it manually when your file's header lies.

Outputs

The useful ones: IMAGE (the frames, as a batch), MASK (alpha from EXR, or generated for other formats), frame_count, width, height, fps - wire those four into your layout and resolution nodes so the rest of the graph adapts to the footage instead of hardcoding - plus audio (as an AUDIO type, pipe it into a video write later) and video (a string carrying the source identifier). That last one matters if you want to pass the source name through to a metadata overlay or a filename.

The honest bits

The pack's launch reception was mixed - community members liked the ambition but flagged the 32-bit claims as overstated in places and the docs as thin. Read is one of the better-behaved nodes: it uses OpenCV's VideoCapture under the hood, which is battle-tested. Where people actually get burned: forgetting frame_limit on a long clip and blowing up RAM, and mis-declaring the input colorspace on log footage. Also note this node appears in the menu under two registration names - with and without the ◎ prefix - they're the same node, so don't install anything twice.

Install

Part of fxtdstudios/radiance - Manager search "Radiance", or:

cd ComfyUI/custom_nodes
git clone https://github.com/fxtdstudios/radiance.git
cd radiance
pip install -r requirements_linux.txt   # or windows/mac_silicon

Restart after; Linux needs libopenexr-dev before pip. OpenEXR, imageio-ffmpeg and OpenColorIO make the install heavy and the first boot slow.