Meld Image Loader (Batch)

The single-image Meld Image Loader is great for one-off iteration. Meld Image Loader (Batch) is its older sibling for when you want to walk through a folder of images in sequence - img2img over a directory, upscaling a pile of PNGs, or reprocessing a whole export. Instead of an image picker, you give it a directory and an index, and it loads the nth file, sorting by filename.

Same metadata superpowers as the single version: it parses the embedded workflow/prompt info and tries to restore the generation settings, including loading the model.

The inputs that matter

• directory_path - the absolute path to the folder (default C:\Images). It loads .png, .jpg, .webp, whatever's there.
• index (default 0, 0-based) - which file to load, from the list sorted by filename.
• stop_at_limit - the interesting one. If false (default), the index wraps via modulo - index 99 on a 10-file folder silently loads file 9. If true, an out-of-range index raises an error and stops the queue.

How to actually batch with it

The trick is making index advance. In ComfyUI, right-click the index widget, convert it to a Primitive input, and set its "control_after_generate" to increment. Then every Queue execution loads the next image in the folder: run once, it's file 0; run again, file 1; and so on. Bump the Batch Count if you want several in one go.

With stop_at_limit false, an oversized count just loops back to the start - handy when you don't know exactly how many files are there and you'd rather cycle than crash. With it true, you get a clean error at the end of the folder, which is what you want for a fully-automated pass where missing files should be loud.

The outputs

Same eight as the single loader: IMAGE, MODEL, CLIP, VAE, positive, negative, summary, and base_settings. So each image in the sequence can reconstruct its own prompt, settings, and model - which makes "re-render every image in this folder with the same tweak" genuinely possible, because each iteration re-reads that file's own recipe rather than one shared config.

Gotchas

Metadata parsing is only as good as the files. Images saved by ComfyUI with embedded workflow JSON parse beautifully; exports stripped of metadata come back with defaults, and the model load will no-op. Check the summary output on the first pass to confirm it's reading what you expect. And keep the folder to the directory you intend - the pack's file loading is scoped to the path you specify and its subdirectories.

Installing it

Same pack, same install:

• ComfyUI Manager → search "Meld" → install, or
• comfy node install HappyOnigiri/ComfyUI-Meld, or
• cd ComfyUI/custom_nodes && git clone https://github.com/HappyOnigiri/ComfyUI-Meld.git && cd ComfyUI-Meld && pip install -r requirements.txt then restart.

No models to download - it loads what's in the metadata. If you're doing a folder-sized job, the increment-on-generate + stop_at_limit combination is the whole game; the wrap-around default will surprise you the first time you run past the end of the list, so pick your mode on purpose.