How do I create a manifest file for batch jobs?

Create a JSON file with an array of job objects. Each object needs at minimum a prompt, model, and output path. Example: [{"prompt": "sunset over ocean", "model": "flux", "output": "./images/sunset.png"}]. Save the file and run gen-ai batch --manifest jobs.json. The CLI validates the structure before starting. Download a sample template at picsart.com/cli.

Can I run batch jobs on hundreds of images at once?

Yes. The CLI handles manifests with hundreds or thousands of jobs. Use --max-jobs to control how many run in parallel (default is 5). Higher concurrency finishes faster but can hit API rate limits or exhaust credits quickly. For very large batches (1000+ files), split into multiple manifest files and run them sequentially.

How do I pipe batch results to other tools?

Add --json to get structured output. Example: gen-ai batch --manifest jobs.json --json | jq -r '.results[].url' extracts all file URLs. Pipe to curl to POST results to an API, or save to a file for later processing. JSON mode is silent (no progress logs), so it's perfect for CI/CD pipelines.

Can I control how many jobs run at the same time?

Yes. Use --max-jobs followed by a number. Example: --max-jobs 10 runs 10 generations in parallel. Lower numbers reduce load on your account and API limits. Higher numbers finish batches faster. Start with 3-5 for testing, then scale to 10-20 for production runs.

How do I upload batch outputs to Google Drive or Dropbox?

After your batch completes, run gen-ai drive sync --directory ./results to upload the entire output folder. The CLI connects to your Drive or Dropbox account (one-time auth) and syncs all files. Add --watch to keep the CLI running and auto-upload new files as they're generated.

Is there a limit to how many files I can process in one batch?

No hard limit on manifest size, but practical limits exist. Very large batches (5000+ jobs) take hours to process and consume significant credits. Split into smaller manifests (500-1000 jobs each) for easier debugging and cost tracking. Your account's credit balance is the real limit — run gen-ai estimate on your manifest to check total cost before starting.

Can I use batch mode to re-render product photos with new backgrounds?

Yes. Create a manifest where each job points to a product image with --image and includes a prompt like "clean white background." The CLI processes all images in parallel, applies background removal or replacement, and saves to your output directory. Perfect for catalog re-shoots or seasonal campaign updates.

How to run batch AI generations and script workflows with the CLI

SKILLSAdvanced

What you'll learn

What is batch generation with the CLI?

Common use cases

Run batch generations step by step

STEP 1: Create a manifest file

On web: Go to picsart.com/cli → Download a sample manifest template
On mobile: Not applicable — batch jobs run in terminal only

Get sample manifest

STEP 2: Configure batch settings

Set up your manifest and concurrency options:

Manifest format: JSON array with prompt, model, and output path for each job
Concurrency: Add --max-jobs 5 to run 5 generations in parallel (adjust based on your account limits)
Resume mode: The CLI skips files that already exist, so you can re-run after failures
Output directory: Use --output ./results to save all files to a specific folder

STEP 3: Run the batch job

Execute gen-ai batch --manifest prompts.json with your manifest file. The CLI processes all jobs in parallel based on your concurrency limit. Progress updates show how many jobs are complete, pending, or failed. Outputs save to your specified directory.

STEP 4: Review and verify

Check the output directory for completed files: Not perfect? Adjust prompts in your manifest and re-run. The CLI skips already-generated files and only processes new or failed jobs.

Verify all expected files were created
Check for failed jobs in the CLI output log
Inspect a sample of outputs to confirm quality meets your standards

Start batch processing

Tips for best results

💡 Start with low concurrency and scale up

Begin with --max-jobs 3 to test your manifest structure and verify outputs. Once you confirm everything works, increase to --max-jobs 10 or higher. Too much concurrency can hit API rate limits or exhaust your credit balance faster than expected.

💡 Use JSON output for piping to other tools

Add --json to get structured output you can pipe to jq, curl, or CI scripts. Example: gen-ai batch --manifest prompts.json --json | jq -r '.results[].url' extracts just the file URLs. Perfect for posting results to Slack, uploading to S3, or triggering downstream automation.

💡 Resume failed jobs without re-generating

If a batch job fails midway, re-run the same command. The CLI checks your output directory and skips files that already exist. Only failed or pending jobs get processed. This saves time and credits when handling large batches.

💡 Organize manifests by project or campaign

Keep separate manifest files for different projects: product-catalog.json, social-campaign.json, blog-images.json. This makes it easier to re-run specific batches, track credit usage per project, and version-control your generation workflows.

Manifest structure reference

Frequently asked questions

Ready to automate?

Install the CLI and start running batch jobs to process hundreds of generations in parallel.

Get started