Faishme Repeat Image Batch

Faishme Repeat Image Batch takes an image batch and repeats each individual image a set number of times, in place. The distinction matters and trips people up: a batch of [A, B] with amount 2 becomes [A, A, B, B], not [A, B, A, B]. Each row is duplicated back-to-back, so the ordering you had is preserved while the batch gets fatter. It's a batch-shaping tool, not a generator.

How it works

The implementation is a tensor reshape-and-repeat, but the intuition is simpler than the math. The node reads your batch as (batch, height, width, channels), inserts a repeat dimension, tiles each row amount times, and flattens it back down. Net effect: every image now appears amount times consecutively, and the batch size multiplies by amount. A 4-image batch with amount 3 → a 12-image batch. The amount input runs 1 to 4096, so it'll happily blow up a small batch into something enormous if you let it.

Why you'd reach for it

This is a batch-expansion utility, and the reason it exists in a fashion pipeline is straightforward: you have a folder of model shots and you want every shot fed through the next stage several times - with a different seed, a different pose prompt, or a different garment - before collecting results. Repeat each image N times upstream, vary something per-position downstream, and you've got N variations of every source without reorganizing your data. It's the same family of trick as the pack's Stack/Unstack nodes, just on a single axis.

Inputs and outputs

• image - the IMAGE batch to expand.
• amount - how many copies of each image, 1 to 4096 (default 1, which changes nothing).

Output is a single IMAGE batch, amount× the input size.

Installing it

Standard for the pack - ComfyUI Manager, search "ComfyUI_faishme", or:

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

Restart ComfyUI. It shows up under FaishmeNodes, and needs nothing beyond the usual environment.

Where people get burned

Two things. First, the order semantics - if you actually wanted [A, B, A, B] (interleaving the whole batch), this node does the opposite and you'll need a different approach. Second, memory: multiplying a big batch by a big amount is a silent VRAM bomb, and since it happens before whatever computes on the batch, the OOM can look like it's coming from the next node. Keep amount modest unless you know the downstream stage can eat the expanded batch. Otherwise it's a clean, predictable utility - the kind of tiny node that makes batch pipelines work without a custom script.