Bulk Generation with Wildcards

Bulk generation lets you run dozens (or hundreds) of variations of the same prompt without sitting at your keyboard. A background worker claims jobs from the queue and generates them on a schedule, billing your credits as it goes.

When to use it

• Exploring style variations ({watercolor|oil|pastel})
• Generating a product catalog with many subjects
• Building training-style data sets
• A/B testing prompts across multiple subjects

Setup

There’s no bulk mode toggle. Bulk runs are auto-detected the moment your prompt contains a {a|b|c} wildcard. Just write the prompt and click Generate as usual — Banana Flow routes it through the bulk pipeline automatically.

1. Add a Prompt node. Use {a|b|c} syntax for any axis you want to vary:

   A {watercolor|oil|charcoal} portrait of a {cat|fox|owl} in a {forest|library|cafe}

   That single prompt expands to 27 combinations (3 × 3 × 3).

2. Connect the prompt to an Output node as you would for any generation. Optionally connect source images.

3. Pick a model and aspect ratio on the Output node.

4. The Output node shows a small chip below the prompt: → N prompts · X credits — that’s the live count and total cost.

5. Click Generate (the normal button). Banana Flow expands every wildcard combination, charges per-combo as it runs, and queues the jobs.

Wildcard syntax

• {a|b|c} — pick one of these literal options
• \{ and \} — escape literal braces in your prompt
• Nested wildcards ({a|{b|c}}) are not supported

How it runs

A cron-driven worker picks jobs out of the queue every minute and generates them one at a time. You can close the tab — generation continues in the background. Results appear on the Output node as they arrive.

Watching progress

Open the History panel (the floating “History” button on the top-right of the canvas). Bulk runs appear as their own cards with:

• Completed / total counts
• Per-job status as it streams in
• A Cancel button to stop the rest of the queue

You can close the tab — the cron worker keeps draining the queue server-side either way.

Costs

Total cost = (number of combinations) × (per-generation credits for the chosen model). It’s charged per job as it runs, so failed generations are refunded just like single-shot ones.

Tips

• Start small (2–3 axes, ~10 combos) to validate the prompt format before committing 100+ generations.
• Use the cheapest model that gives acceptable quality for bulk jobs (Nano Banana 2 is usually the right call). Reserve premium models for finishing passes.
• Turn on auto-refill in Billing if your bulk job will outlast your current credit balance.