AM Read Image

Stock LoadImage reads PNGs from ComfyUI/input/ and calls it a day. AM Read Image is the node you reach for the moment your source material stops being PNGs in a folder and starts being EXR/DPX/TIFF sequences on a render drive. OpenImageIO-backed, it reads basically any image format a studio produces, understands frame sequences, color-manages them into your working space, and splits alpha out the way ComfyUI expects.

It's part of comfyui-am-vfx-tools ("AM VFX Tools" category), Adrian Meyer's 13-node VFX toolkit - the public slice of an internal studio pipeline. The OIIO read path started from sumitchatterjee13/nuke-nodes-comfyui and was rewritten substantially, which is why it smells like a real I/O node rather than a toy.

How it works

The core idea is a frame-token path: instead of loading one file, you give it a pattern and it expands it. All three VFX conventions work:

/work/shots/sh010/plate.0001.exr        # single frame
/work/shots/sh010/plate.####.exr        # 4-digit zero-pad
/work/shots/sh010/plate.%05d.exr        # printf style
/work/shots/sh010/plate.$F4.exr         # Nuke style

Then a frame_mode decides what to load: single (just first_frame), range (a window, with before/after edge policies of hold/loop/bounce/black), or all (every frame on disk). There's a missing_frames policy for gaps - error, black, checkerboard, hold, or nearest - which is the difference between a render failing and a render limping through a dropped frame. The 🔍 Detect Range button next to the frame fields scans the directory and auto-fills the window.

The inputs that matter

• file_path - the path or sequence pattern. Use the 📂 Browse button; it opens the OS-native file dialog.
• input_colorspace / working_colorspace - the OCIO transform from the file's tagged (or assumed) space into the space downstream nodes see. raw_data skips the transform entirely if you want pixels verbatim.
• frame_mode + first_frame/last_frame - what to load.

Everything else - reformat-to-box, scale, filter, and an output_dtype (fp16 default, half the RAM; flip to fp32 if a downstream node complains) - is convenience baked into the read.

Outputs

image (RGB, N×H×W×3), mask (alpha, following the stock convention where mask = 1 - alpha, so white = transparent/inpaint area - wire a MaskInvert if you want Nuke-style natural alpha), resolved_path, info, width, height, frame_rate, frame_count, and a video socket. That last one is a VideoFromComponents convenience wrapper around the already-decoded batch - no RAM benefit, it's just there so you can feed stock video nodes. The real RAM trick is on AM Read Video's video socket, which is a true file reference.

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 openimageio dependency (imports as OpenImageIO) comes from that pip line.

Where people get burned

The frame token. Forget the #### and type a literal plate.0001.exr, and you load one frame and wonder why the sequence "isn't a sequence." Also: missing frames default to black, which silently inserts black frames into your batch - flip it to error if you'd rather a render fail loudly than ship a black hole. And if your existing ComfyUI environment wasn't built with EXR support, set export OPENCV_IO_ENABLE_OPENEXR=1 before launch - the pack sets it in-process, but an earlier custom node that touched OpenCV may have already committed to the default.