Faishme Unstack Images

Faishme Unstack Images is the teardown half of the pack's batch pair. Where Stack Images merges several sources into batched tensors, this node takes batched images and splits them back into separate tensors, one per image. You stack before the sampler to generate in a batch, then unstack after to handle each result individually - that's the workflow shape this node exists to complete.

How it works

The mechanism is a chunking split. For each incoming batched image, the node slices it along the batch dimension into consecutive chunks of stack_size and emits each chunk as its own entry in the output list. A batch of four images with stack_size 1 becomes a list of four single-image tensors; with stack_size 2 it becomes two tensors of two. Like its sibling, it's list-aware (INPUT_IS_LIST), so it accepts multiple incoming batches and splits each one.

Inputs and outputs

• image - the batched IMAGE (or several, as a list) to split.
• stack_size - images per output chunk, 1 to 16.

Output is a single image - a list of tensors, each holding up to stack_size frames. It's marked as an output node, so it can terminate a branch, though feeding the split images onward is the actual point.

Why you'd use it

Post-batch fan-out. Samplers love batches; per-image processing doesn't. So the efficient pattern is: gather images with Stack Images → generate as a batch → unstack → run each image through a per-image step like a detailer, a VAE decode you want separately, or a save that needs one file per frame. In the Faishme pipeline specifically, that's how a folder of model shots becomes a batch, gets processed, and comes back out as individual images ready for per-shot saving with Faishme Save Image.

Installing it

The standard pack install - ComfyUI Manager, search "ComfyUI_faishme", or:

cd ComfyUI/custom_nodes
git clone https://github.com/AkashKarnatak/ComfyUI_faishme

Restart ComfyUI. It appears under FaishmeNodes.

Where people get burned

The usual list-semantics confusion applies in reverse: this splits batches, not sources, so wiring one single image into it with stack_size 1 just gives you back that one image. The real trap is chunk alignment - a batch of 5 with stack_size 2 leaves a ragged last chunk of 1, which some downstream nodes quietly accept and others choke on. If your batch size isn't a clean multiple of stack_size, normalize it upstream or you'll be debugging an off-by-one that shows up one node later. Used correctly, it's the tidy bookend to Stack Images.