Doom Image Loader

Doom Image Loader is the pack's batch image loader, and the clue to its real purpose is the two-stream design. Beyond the stock "load one image by name" behavior, it reads from folders, applies filters, and - the distinctive part - drives two independent streams (A and B) so you can pair images from two folders and step through them in lockstep. If you've ever hand-built a "feed every image in this folder through the pipeline" graph, this is the node that makes it a config setting instead of a graph.

It's the front end of the pack's small data-utility family (Loader → Router → Resize), aimed at batch workflows: run a folder of source images through an img2img or reference pipeline, or generate a contact sheet, or build a comparison loop where stream A is the source and stream B is the style reference.

How it works

Three modes cover the layouts:

• single_folder - one folder, stream A only.
• dual_parallel - two folders (base_path + second_path), streams A and B stepped independently, each with its own index.
• subfolders_seq - recursively walks subfolders of base_path in sequence.

Selection is sequential (walk the index), random (new random pick each run), or shuffle_once (random with a stepped index). index_a / index_b are the current positions - bump them to move through a folder - and reset_index zeroes both per run.

The filters are where it gets genuinely useful: name_contains (substring), min_width/max_width/min_height/max_height (0 = off), min_aspect/max_aspect (w/h ratio range), and orientation (landscape/portrait/square). Combined with extensions (default *.png;*.jpg;*.jpeg;*.bmp;*.webp, ;-separated) and sort_by (name or date), you can narrow "all files" to "landscape PNGs between 1024 and 2048 wide, newest first" without a single line of code.

Outputs: image_a + meta_a and image_b + meta_b. The meta dicts carry filename/width/height/aspect/index/total/source - and that's the handshake with the pack's DoomImageRouter, which routes images based on exactly those fields.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/PeterMikhai/Doom_Flux_NodePack

Restart or use ComfyUI Manager. No extra deps; current ComfyUI (V3 API). The README's DoomAI_nodes.git line is stale - repo is Doom_Flux_NodePack.

Common issues

• Empty outputs, no error. Almost always a path or filter problem: base_path must be an absolute path that exists, and any 0-vs-set filter confusion is worth a double-check - min_width of 0 is "no filter", but a forgotten max_aspect of, say, 1 will silently drop every image wider than 1:1.
• Sequential mode loops forever - it wraps with index % length, so a folder of 3 images cycles 0→1→2→0. That's by design; use it as a loop, don't fight it.
• Two streams out of step - each stream has its own index; if A and B are supposed to advance together, set reset_index and drive both indices from the same counter (the pack's DoomIndexPairCounter is built for exactly this pairing problem).
• Big folders are slow on first scan - the size/aspect filters open every candidate file with PIL to check dimensions, so a 10k-image folder plus dimension filters means a real delay on the first pass.

It's a utility node, but it's the utility node that makes "run my whole dataset through Flux.2 references" a two-minute setup instead of a small programming project.