BMAB Load Output Image

Small quality-of-life node, and once you've used it you'll wonder why core ComfyUI doesn't have one by default. Load Image reads from your input/ folder - which means every time you want to take a previous result and feed it back into the graph for another pass (upscale it, inpaint over it, run it through BMAB Reframe), you first have to manually copy the file from output/ into input/. This node just points at output/ directly.

How it works

Same mechanism as Load Image, different folder. The dropdown lists whatever's actually in your ComfyUI output directory - the schema's own sample value is literally a placeholder, _output_images_will_be_put_here, which shows up when you haven't generated anything yet. Once you've run a few generations, that dropdown fills with your actual output filenames and you pick the one you want to continue from.

The inputs and outputs that matter

• image (enum, required) - a dropdown of files from your ComfyUI output/ folder. Empty until you have output to point at.
• Outputs: IMAGE and MASK - same pairing as any image loader; the mask comes from an alpha channel if present.

That's the whole node - one input, two outputs, no settings.

How to install it

ComfyUI Manager: search comfyui_bmab. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/portu-sim/comfyui_bmab.git
cd comfyui_bmab
pip install -r requirements.txt

Requires no model downloads and doesn't depend on BMAB's ControlNet or IP-Adapter companion packs to work - this is filesystem plumbing, not a generation node.

Common issues & troubleshooting

Dropdown is empty, and you've definitely generated images. Check that ComfyUI's configured output directory (the same one your Save Image node is writing to) is what this node is reading from - if you're running with a custom --output-directory flag or a mounted volume, make sure it's actually pointing at the same location. On a hosted/serverless setup like ComfyICU, also confirm you're looking in the same run's output rather than a previous session's, since output can be scoped per-run.

You picked an image, but it's an old one you forgot about. Because this browses the whole output folder rather than "the last thing you generated," it's easy to grab an older file by mistake if filenames are similar (e.g., sequential ComfyUI_00001_, ComfyUI_00002_...). Sort or filter by date in your file browser first if you're not sure which one is current, or use distinct filename_prefix values on your Save nodes so it's obvious at a glance.

Iterating on the same image feels slower than it should. This node is exactly the fix for the manual copy-to-input workflow, but remember it only reads - it doesn't write. Pair it with BMAB Remote Access And Save or core Save Image at the end of whatever pass you're running, so the next iteration has something new to load.