Finally got tired of manually regenerating 50+ UI icons every time the design token specs drifted. Leonardo's batch features looked promising, so I ran it through a real stress test.
The API is straightforward. You feed it a CSV of prompts and parameters, kick off a job, and poll for results. Works... until your dataset isn't perfectly clean. Learned the hard way: their queue system hates malformed entries and fails the entire batch silently. No partial completions. 😑
```bash
# Basic job submission curl. Note the 'urls' field for custom models.
curl -X POST
-H "Authorization: Bearer $LEO_KEY"
-F "csvFile=@/path/to/batch.csv"
-F "params={"modelId":"e71a1c2f-..."}"
https://cloud.leonardo.ai/api/rest/v1/batches
```
Key takeaway: pre-process your prompts like your CI/CD depends on it. Validate lengths, strip weird characters. Treat it like a production pipeline, not a playground. Got my icon set rebuilt in under an hour once the inputs were sanitized.