Batch Image Upload

The name promises a folder of images turned into a ComfyUI IMAGE batch. The code delivers a shrug.

Let me be straight with you before you build anything on this one: as of the current commit, BatchImageUpload does not actually load images. The input schema declares a source_dir string, but the function behind the node never reads it - it expects an images tensor that it never declares, and just echoes that tensor straight back out. When ComfyUI executes the node it calls run(source_dir=...), the signature says run(images), and you get a TypeError at queue time. That's not a setup problem on your end. It's an unfinished scaffold wearing a finished name.

That's a shame, because the intent is legible. This pack is built around captioning - feed it a folder of images and the Ollama captioning half turns them into text files. A folder-to-batch loader is exactly the glue that pipeline wants, and having it inline would save you wiring one Load Image node per file. But "wants" is doing a lot of work there. Right now the node can't do it.

So what do you actually do? For today, skip it. If you need a folder of images inside the graph, the built-in Load Image node per file still works, and there are mature packs with real folder loaders if you're doing this at scale. If you're handy with a Python node, loading a directory and returning an IMAGE batch is genuinely about ten lines - you'll finish it faster than this author did. The good news: the node is small, self-contained, and isolated. Its brokenness doesn't take the rest of the pack down with it. The llama.cpp, Ollama, and caption nodes are separate classes with separate code.

One thing worth saying about this pack while you're here: it's brand new (created December 2025), has an empty README, and the author "LatentGEN" has no community footprint to speak of. None of that is a crime - every pack starts somewhere. It just means you shouldn't treat any of its nodes as settled. Check for an updated commit before you wire something permanent to a node this green. This one in particular: it's the pack's weakest link right now, and the most likely to be fixed (or renamed) next.

Installing the pack

ComfyUI Manager is the easy path: open Manager → Custom Nodes Manager → search ComfyUI-LLMNodes and install. Or clone it by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/LatentGEN/ComfyUI-LLMNodes

Then restart ComfyUI. The pack declares zero Python dependencies in its pyproject.toml and ships an empty requirements.txt, so there's nothing extra to pip install for the utility nodes - you'll only hit that with the llama.cpp nodes.

Troubleshooting

• "TypeError: run() got an unexpected keyword argument 'source_dir'" at queue time - expected, given the current code. Not you, not your model paths.
• Node exists but does nothing useful - also expected. There's no workaround inside the node; swap in a real image loader.

Keep an eye on the repo. When the author actually reads that directory and emits an IMAGE batch, this becomes a handy node. Until then, treat it as a placeholder and don't build a workflow on it.