When you need thirty pictures rather than one, bulk generation runs them as a batch. You set it going, leave, and come back to the results.
When it is worth it
- Many variations of one idea, to choose between
- The same treatment applied to a list of subjects — a product range, a cast of characters
- A large set for a single purpose, where you will discard most of them
If you only want a handful of variations of one prompt, you do not need this. The how many setting on the normal panel already generates several at once — see settings explained.
Before you start a batch
Get the prompt right on single images first. A batch multiplies whatever you give it, including the mistakes, and a hundred images built on a prompt that was not ready costs a hundred times as much to discover it.
The habit that saves the most credits: run one, look at it properly, then run the batch.
Running one
From the bulk area of your workspace, set up the run — the prompt or list of prompts, the settings, and how many. The total cost is shown before you start, and it is the sum of every item in the batch.
Then start it and leave. Batches take a while by nature, and nothing needs you present.
While it runs
A batch has its own page showing progress: what is finished, what is still going, what failed. Finished items appear as they complete rather than all at the end, so you can start judging early — and cancel early if the first few show the prompt is wrong.
Notifications can tell you when a batch is done. See notifications.
When part of a batch fails
A batch does not stop because one item failed. The rest continue, and the batch finishes in one of these states:
| State | Meaning |
|---|---|
| Completed | everything worked |
| Partial success | some worked, some failed — the successes are yours to keep |
| Failed | nothing usable came out |
| Cancelled | you stopped it |
Partial success is normal on large batches and is not a sign anything is wrong with your account.
Credits are only spent on work that produced something. Items that failed do not consume credits — the amount held for them is released back to your balance. See when something goes wrong.
Where the results go
Into your library like anything else, grouped so the batch stays together rather than scattering through Recent.
Sensible next steps:
- Go through the batch and favourite the keepers.
- Delete the rest while you still remember which is which.
- Move the keepers into a folder.
Doing this at the end of each batch takes two minutes. Leaving it until you have run five batches turns it into an afternoon.
Sharing a batch
A finished batch has its own address you can send to someone — useful for showing a client a set of options in one link rather than sending files.
Fair queueing
Large batches are interleaved with other people’s work rather than running as one uninterrupted block. This is why a hundred-item batch does not finish in exactly a hundred times the length of one item, and why someone generating a single picture is not stuck behind your batch.
The trade is deliberate: it keeps the queue usable for everybody, including you when you are the one making a single picture.