Frequently asked questions

No — contractually and architecturally. Every licensee’s corpus, corrections and calibration profile are isolated per tenant. Your data never improves anyone else’s drafts, and no shared model is trained across shops.

If you end your licence, your calibration profile leaves with you.

You do. Drafts generated under your licence, and the finished files your experts approve, are your work product — the same as if your team had digitized them from scratch. We charge for the drafting service, not for rights in the output.

Useful calibration starts at around 25–50 production files — our published fingerprint was measured on 26. More files sharpen it: with a few hundred, the profile resolves per technique and design size; with a full back-catalogue, per fabric and garment type.

The most valuable files are ones paired with their original artwork, because each pair lets us measure your decisions against the source — but stitch files alone carry most of the signal.

Those tools ship one generic algorithm for every user, which is why their output never looks like your work. StitchDraft is calibrated on your own back-catalogue — your densities, your satin widths, your underlay and trim habits — and then keeps learning from your experts’ corrections on live jobs. Generic tools are frozen; yours compounds.

Strong today: left-chest-scale logo work — satin lettering, tatami fills with underlay and pull compensation, fine-detail run stitching, multi-colour sequencing.

Routed to your experts: wide ornate serif lettering, very fine detail near needle limits, 3D puff, appliqué, caps, and large-format jacket backs. The workflow flags these rather than pretending; your team digitizes them manually as they do now — and those files feed the next calibration pass.

PNG, JPG, PDF and vector formats (AI/EPS/SVG via PDF). Flat logo artwork digitizes best. Photographs and heavily gradiented art are flagged for human handling — that’s true of manual digitizing too.

White-on-transparent artwork (logos destined for dark garments) is handled correctly — a common failure in generic tools.

Native output is .DST (Tajima) — the universal production format every commercial machine and every major digitizing suite reads. Conversions to .PES, .EXP, .JEF and other machine formats are included.

We don’t write .EMB — that’s Wilcom’s sealed proprietary format. Wilcom imports our .DST directly, which is the standard industry route.

Review happens in the browser: sewn preview, stitch counts, density map, approve or send back. For touch-ups, your team keeps using the software they already know — corrected files are simply uploaded back, and the engine learns from the diff automatically. No new digitizing suite to learn.

A draft your team approves for production — as-is or after touch-up. Drafts, regenerations and rejected files cost nothing. If a draft is so far off your expert digitizes from scratch, you don’t pay for it; you only pay when the engine actually saved the work.

It replaces the mechanical 80% of the job, not the judgement. Shops running StitchDraft point the same team at several times the volume: the engine drafts, experts approve and fix the genuinely hard work. Faster turnaround becomes something you sell; your experts spend the day on work worth their skill.

What it does change: growth stops requiring you to find and hire scarce digitizers.

Two weeks, no commitment: you provide ~100 backlog files for calibration, then run live jobs through the engine alongside your normal process. We measure three numbers together — acceptance rate, minutes-to-fix per draft, and how fast both improve. You keep every file produced. If the numbers don’t convince you, you walk away with a free style analysis of your own catalogue.

Pricing is agreed with founding partners as part of the pilot, and the model is simple: you only pay for accepted files — drafts, regenerations and rejected files cost nothing, and there are no seats or minimums. Early shops lock in preferential founding-partner terms.

Get in touch and we’ll talk numbers against your actual volume.