Batch Latents (DEPRECATED)

Straight to the point: this node's display name in the UI is literally "Batch Latents (DEPRECATED)". It stacks two latents into one batch, and it still works, but ComfyUI ships a better replacement, so you should use that one for anything new. This article is mostly so you can recognize the old node in someone else's workflow and know what to swap it for.

What it did

LatentBatch took samples1 and samples2 and joined them along the batch axis, producing one LATENT containing both. That's the entire feature set. The batch axis is the one that makes a sampler process multiple latents in parallel - which is how you generate a grid of variants in a single sampling run instead of looping the graph. It combined two and exactly two inputs, which was always the limitation.

Why it's deprecated

The job hasn't gone away, the node just got strictly better. The replacement is Batch Latents (BatchLatentsNode), which does the same stacking but is autogrow: it takes 1 to 50 latent inputs rather than a hard-coded two. Batch two, batch ten, batch forty - same node, no rewiring. The old two-input version survives in saved workflows and muscle memory, and ComfyUI keeps it registered rather than deleting it so old graphs don't break, but it's frozen in place and won't grow.

The gotcha that trips everyone

Batching isn't concatenating. LatentBatch stacks along the batch dimension - two latents of equal shape become one batch of two. It does not join along x/y/t; that's LatentConcat's job. Get the two mixed up and you'll either get a shape error or, worse, a "batch" that's actually a stitched image. The rule of thumb: batch = more samples in parallel, concat = bigger single sample.

Also worth knowing: when you batch latents, the sampler processes them as independent denoising runs, so each keeps its own content - but the batch shares a single seed unless you randomize per item. If your two batched latents are meant to be different, make sure their noise sources (or the node feeding them) differ; otherwise you get the same result twice, which is the classic "why are my batch outputs identical" complaint.

Should you use it?

New graph? Use BatchLatentsNode. You inherited an old workflow with this node and it works? Fine, leave it alone - it's deprecated, not broken. Just don't build anything new on it, and when you bump into it in someone's tutorial from 2024, you'll know exactly what it is and what replaced it.