Building a batch image workflow that does not waste your evening
The short answer
Set up the output contract first (naming + folder layout + format), then run one preset over a 20-file sample, verify the manifest, and only then process the full batch. Doing it in that order turns an overnight job into a four-step routine you can repeat.
A practical checklist for setting up repeatable bulk image processing: naming, presets, verification and rollback.
- #workflow
- #images
Building a batch image workflow that does not waste your evening
Batch processing thousands of images is easy to start and painful to undo. This checklist is what we use before running any large job.
1. Decide the output contract first
Before touching a single file, write down:
- target format and quality
- naming pattern (prefix, index padding, extension)
- destination folder — never write into the source folder
2. Build a preset, not a pile of flags
A preset is a saved, named configuration. It makes the job repeatable and, more importantly, reviewable by someone else.
preset: web-gallery
format: webp
quality: 82
maxWidth: 2400
naming: {prefix}_{0000}
3. Run on a sample, then verify
Always run the job on 20 files first, then check:
- dimensions and file sizes are in the expected range
- EXIF/orientation is preserved or intentionally stripped
- the naming pattern actually produces unique names
4. Keep a rollback path
The cheapest rollback is not overwriting: write to a new folder and keep the originals until the job is verified. Disk is cheaper than your evening.
FAQ
- How many files should the sample run include?
- Twenty is enough. It surfaces naming collisions, rotation problems and colour-profile surprises while the batch is still cheap to redo.
- Should I keep the originals?
- Yes, until the verification pass is done. Keep them in a separate folder and only delete after you have compared a few outputs side by side.
- What breaks most often?
- Filenames with spaces or non-ASCII characters, and images that carry EXIF rotation. Fix both before scaling up.
References
Related articles
- Running vision models on your own machine: the four costs nobody mentions
Local AI · Sep 14, 2026
- How to evaluate an AI tool before you trust it with real work
AI Tools · Sep 12, 2026
- Ollama, in plain words: what you can do with an AI model that runs on your own laptop
Open Source · Sep 19, 2026
Product mentioned
Stack Pix
Node-based AI photo studio that runs entirely on your PC