Image Queue Processor

Image Queue Processor is the image-side version of aidec's Text Queue Processor: hand it a batch of images and it serves them to your sampler one at a time, each a full separate img2img run, until the batch is done. Same serial-queue idea that makes the pack useful - process a stack of images without cramming them all through one giant tensor.

Here's the honest catch, and you should know it before you wire anything: this node doesn't read a folder. It takes an already-loaded images batch as input, which means you need a separate loader in front of it - the author points specifically at Inspire Pack's "Load Image Batch From Dir" or similar. And because that loader pulls the whole folder into memory up front, this approach gets heavy fast. A couple hundred images can sit on several gigabytes of RAM before a single generation happens.

Which is why, for batch img2img, the newer Image Queue Processor Pro is the one the author now recommends - it reads the folder itself and lazy-loads by index, keeping memory tiny. Think of this original node as the still-functional predecessor: fine for small stacks, outgrown by big ones.

How it works

The engine is the same server-side counter the whole pack runs on. Each queue, it emits the image at current_index from the incoming batch, advances the counter, and - being an output node with trigger_next on - re-queues the workflow so the next image runs automatically. It walks the batch, completed flips true at the end, and it stops.

The inputs that matter

• images - the batch to iterate. This comes from an upstream loader (Inspire's Load Image Batch From Dir is the usual pairing); it's not a folder path.
• start_index - where to begin (0 normally).
• trigger_next - auto-advance; on drives the whole batch, off gives you one image per manual queue.

Outputs: image (the current one, into your VAE Encode / sampler), plus current_index, total, completed, and status. That current_index is worth routing into a filename node - pair it with Image Filename Processor and the pack's dir loader to keep output names matched to inputs.

Installing it

cd ComfyUI/custom_nodes && git clone https://github.com/aidec/Comfyui_TextBatch_aidec, restart ComfyUI. (ComfyUI Manager → Install via Git URL works with the same link.) No models or heavy deps for this pack. It's under the TextBatch category. If you're going to feed it from Inspire Pack, you'll want that installed too.

Where people get burned

• Memory. This is the real one. The upstream batch loader holds every image in RAM at once - fine for 20, painful for 500. If you're doing serious volume, switch to Image Queue Processor Pro instead; it exists specifically to solve this.
• It needs a loader. No directory input here - if you were looking for "point at a folder," that's Pro, not this.
• First run may do just one. Pack-wide quirk; queue again and it continues.
• Reset start_index to 0 between full runs, and keep to a single ComfyUI tab - the counter is shared server state and multiple windows corrupt it.

Not a bad node, just superseded. For anything bigger than a small stack, go Pro.