Batch generation: up to 10 variations in one click

Finding the right prompt always takes a series of attempts: run it, wait, look, tweak, run it again. Now that queue of single generations can be replaced with one click: a multiplier sits next to the generate button, and the service runs several variations of the same prompt at once.

Images go up to ×10, video and music up to ×5. This post covers how it works, how the cost is calculated and why a failed variation doesn't burn your tokens.

How it works

Choose a multiplier from ×1 to ×10 (up to ×10 for images, up to ×5 for video and music), write the prompt once and run it. All variations are generated in parallel, and in your history each one is labeled ×i/N — for example, ×3/10 — so you can see which batch a result came from.

The main use case is prompt tuning: instead of running single generations one after another, you get ten interpretations in one go and see right away which wording works.

What it costs

There's no surcharge for batches: you're charged once — N × the price of a single generation. For example, 10 Nano Banana variations cost $0.33 on the Basic plan ($5 = 150,000 tokens).

If a single variation in the batch fails, its tokens are refunded automatically — the refund covers exactly the failed variation, and the rest keep generating as if nothing happened.

How to try it

Open the generation page, choose a model and find the multiplier next to the run button. Set ×5 or ×10, run it, and compare the variations in your history by their ×i/N labels.