Batch

Several pictures in one run, and how that differs from queuing several runs.

In plain words

A batch is a stack of pictures processed together. Every IMAGE and LATENT in ComfyUI is a batch, usually of one. Set batch_size 4 on the empty latent and the sampler makes four pictures at once, each from different noise, from the same prompt and settings.

There are two ways to get several results:

  • Batch (batch_size > 1): one run, all pictures computed together. Faster per picture on a strong card, but memory grows with the batch.
  • Queue several runs (the Run button's count in ComfyUI, or press Run several times with randomize seed): one picture at a time, same memory as one.

Where you meet it in a graph

  • batch_size on Empty Latent Image, EmptySD3LatentImage and others.
  • Save Image saves every picture of a batch, numbered.
  • Latent From Batch, Image From Batch, Repeat Latent Batch, Rebatch pick out or repeat items.
  • In video, the frames of a clip are one IMAGE batch after VAE Decode; "batch size" of video means several clips and is rarely used.

Values and what they change

  • batch_size 1 to 4 for pictures on a typical card. Video: 1.
  • All pictures of a batch share one seed value; each still gets different noise. To repeat picture 3 of a batch alone, use Latent From Batch or note that its noise depends on the batch.

Typical mistakes

  • Out of memory with a large batch: lower it, or queue runs instead.
  • Expecting different prompts in one batch: a batch shares one prompt (packs exist for prompt lists).

How the app helps

The Empty Latent Image description explains batch_size. Short definitions of this and other words are in the Glossary.

See also

This page is imported from the application repository's wiki (docs/wiki/concepts/Batch.md, revision b8e3f71). The same guide is built into the app: Help ▸ User guide.

To refresh it, run npm run import and then npm run build in the website folder.