NOTES FROM USE
Notes from using it
Short write-ups of jobs we actually run through Stack Pix. These are workflows, not feature lists — each one is a canvas you can open and change.
ID photos: one batch, printed sheet included
Sep 10, 2026
The ID-photo workflow does cut-out, background swap and the repeat_nup layout in one pass, so a folder of portraits comes out as print-ready sheets. The naming template puts each person in their own subfolder, which matters more than it sounds when you are handing files back to other people.
Product shots: cut out, re-background, compress
Sep 5, 2026
Chain cut-out → background → compression and run it over a whole catalogue. The node cache is the part you notice: changing one parameter only recomputes that node and its children, so tuning the result on a big batch does not mean reprocessing everything from scratch.
Culling a trip before editing it
Aug 28, 2026
Run ai_cull first and let it split the set into keep/reject on exposure, focus, closed eyes and duplicates. Colour work goes faster on the survivors, and the reject branch tells you why each frame was dropped instead of silently deleting it.
Old scans: upscale first, restore second
Aug 20, 2026
Order matters. A 4× upscale before face restoration gives the restoration model far more to work with; doing it the other way round just enlarges the blur. Large scans are slow on CPU, so this is the one job where a CUDA GPU changes the experience.
Guides and articles
- Running vision models on your own machine: the four costs nobody mentions
Local inference is not free. It costs disk, memory bandwidth, VRAM and time-to-first-result — and only one of those has anything to do with raw compute.
- Building a batch image workflow that does not waste your evening
A practical checklist for setting up repeatable bulk image processing: naming, presets, verification and rollback.
Stack Pix