Batch PSD to PNG: codify alpha policy, trim padding, and non-silent outputs so a few bad files do not hide in thousands
`batch-psd-png` targets SKU walls, game drops, and localization packs. Failures hurt most when silent overwrites or mixed alpha policies slip into the CDN. Templates must declare which categories need alpha, what trim tolerance preserves shadows, and how manifests record hashes. Automate histogram checks on alpha and file-size outliers; humans spot-check buckets instead of opening every icon.
Como usar
- Pilot ~50 files per category, validate trim does not eat glows, freeze parameter JSON in git, then shard jobs by folder with isolated presets.
- After the run, assert output counts match inputs, quarantine outliers, and trip circuit breakers on repeated error codes instead of blind full retries.
- Ship manifests plus signed sampling reports; retries write new paths or version folders—never overwrite successful bytes you may need to audit.
PSD to PNG (batch) FAQ
Batch PNG alpha inconsistent—some opaque?
Mixed templates—some SKUs need cutouts, others need solid plates—cause half the PNGs to mysteriously lack alpha. Encode per-SKU alpha requirements and script channel detection so inconsistent folders surface before ingest, not after shoppers see checkerboard bugs.
Should we palette PNGs for size?
Palette PNGs shrink dramatically but band on gradients and skin; enable only with art-approved diffs and automated visual compares on sensitive crops, not a single glance at thumbnails.
Trim eats shadows—why?
Aggressive trim clips soft shadows; loose trim wastes pixels. Maintain separate trim profiles for hard-edge icons versus product shots with ambient shadow, and pilot ten shadow-heavy masters before full batches.
Retries overwrite silently—how to trace?
Retries should land in attempt-scoped directories with structured logs; silent overwrites destroy the ability to know which byte reached production or to satisfy compliance questions about who changed what.
Vendor bulk PNG—fast intake?
Run validators on resolution, color mode, alpha presence, and per-SKU size quantiles, route outliers to human review, and keep random visual sampling—this catches systemic palette mistakes in minutes instead of post-launch screenshots.