Generating variations

This is the pattern that turns one workflow into a whole set of results. When an input holds more than one item, you decide how those items reach the generator: each on its own, or all together. That single choice is the difference between getting ten images and getting one. You make the choice with the Each / All toggle on the input node.

​

Each: one result per item

In Each mode, every item starts its own generation. This is the default. Ten product photos become ten images; four prompts become four. The results are independent, so you can review them and keep the ones that work.

​

All: items combined into one

In All mode, the items are bundled and sent together as a single generation. Use it when several inputs should shape one result—for example, three brand references guiding a single new image.

​

Which should I use?

• Reach for Each when each item should produce its own result—generating a set, exploring options, or running one prompt across many products.
• Reach for All when the items work together toward one result—combining reference images, or blending several inputs into a single generation.

​

Mix the modes

You can set one input to Each and another to All on the same generator. The Each input decides how many results you get; the All input is applied to every one of them. Ten products in Each mode, with three brand references in All mode, gives ten images—each using all three references.

​

When two sets need combining

If two independent multi-item sources feed the same generator in Each mode, the result is ambiguous—Sprello can’t tell which item should pair with which. It asks you to add a Combinator to define the pairing, switch one source to All, or restructure.

​

Good to know

• A single node can expand to at most 500 items. If a combination would exceed that, Sprello blocks the run and asks you to reduce inputs.
• Generations run a few at a time and the rest queue, so large sets finish in waves.
• Switching a source between Each and All recalculates the result count. Existing results aren’t deleted—switch back and they reconnect, no re-generation needed.

​

Next