Repeat Latent Batch

Somewhere in your workflow, a node is going to demand that two inputs have the same batch size - and the latent you're feeding it has a batch of 1 while the other branch has 16. Repeat Latent Batch is the duct tape for that situation: it takes every latent in a batch and stacks amount identical copies of it. That's it. It doesn't blend, doesn't interpolate, doesn't add noise. It just makes the same thing appear N times.

Two inputs, one output:

• samples - any LATENT.
• amount - how many times to repeat each item, 1 to 64.
• Output is a LATENT whose batch dimension is batch_size × amount.

A batch of 3 becomes a batch of 3, 6, 9... every image duplicated in place, order preserved. The source code literally does a repeat() on the tensor and copies any attached noise mask along for the ride, so masks survive the trip.

Where it fits

Three honest use cases, in order of how often you'll hit them:

1. Matching batch sizes between branches. Nodes like IPAdapter or per-image conditioning apply the same effect to every latent in the batch. If one branch produces 4 latents and the other produces 1, most nodes either error out or silently use only the first. Repeating the singleton up to 4 fixes it cleanly.
2. Same latent, multiple conditions. You want one base latent sampled with three different prompt branches to compare - repeat it 3 times and let each branch's conditioning do its thing, or repeat and batch through a single sampler with different seeds per image.
3. Grids and batch rendering. You generated one latent you love and want 4 subtle variants of it. Repeat, then sample with 4 different seeds.

Don't confuse it with its siblings

This is where people get lost, because the batch family looks like a pile of near-identical nodes:

• LatentBatch / Batch Latents join different latents into one batch. Repeat Latent Batch makes copies of the same one.
• LatentConcat concatenates along the channel dimension, not the batch - that's for stacking conditioning, not images.
• Empty Latent Image's batch_size is not the same thing. A batch of 4 from Empty Latent Image is four independent noise fields that will produce four different images. A batch of 4 from Repeat Latent Batch is four identical latents - identical output unless something else varies, like the seed.

The trap

The most common beginner mistake is using Repeat Latent Batch when they actually wanted four different starting points - then wondering why every image came out identical. That's not a bug: you literally asked for the same latent four times. If you want variation, either set batch_size on your Empty Latent Image node or vary the seed. Repeat Latent Batch is for when you want the sameness and need the shape to match.

Second trap: repeating a latent with a noise mask attached multiplies the mask too, which is usually what you want (same region protected in every copy), but if you're repeating a masked latent and then changing the mask per image, do the mask work after the repeat, not before.