Print on Demand T-Shirt Design & Listing Generator
Pricing
from $320.00 / 1,000 trademark clearance screenings
Score print-on-demand niches, screen every brand and slogan against the US trademark registry, write copy validated against your platform’s own character limits, and generate print-ready transparent artwork. Supports Amazon Merch on Demand, Etsy, Redbubble and Zazzle.
All notable changes to Print on Demand Listing & Design Generator will be documented in this file.
The input was called keywords and labelled "Seed keywords", while the rest of the Actor — the
niche score, the sub-niche ideas, the niche validator — already said niche. One word now:
| Old | New |
|---|---|
keywords | niches |
keywordsFile | nichesFile |
Nothing breaks. Saved tasks, scheduled runs and API calls sending keywords keep working —
the old names are translated automatically. If you send both, the new one wins.
Build 1.0.3 got this wrong: a run sending only
keywordswas silently given the default nichecat mom t-shirtinstead, and billed for it. The run succeeded, so there was nothing to see. Fixed in 1.0.4 — if you ran 1.0.3 with the old field name, check thekeywordcolumn on your rows.
The dataset field is still keyword, so anything parsing your rows or the upload CSVs is
unaffected; it just shows as 🎯 Niche in the Console now. One value did change: clearance rows
that used to report phraseOrigin: "keyword" now report "niche".
When you supplied your own designText for a text-only design, the screen ran against the seed
keyword instead. So a run for "Knead More Coffee" was screened as "cat mom t-shirt" — a search
term that never reaches a garment — and the printed wording was never checked at all.
Now the printed string is what gets screened and what the design is gated on. If it comes back
blocked, no artwork is generated and the row says which phrase was blocked. Clearance rows carry
phraseOrigin: "designText" so you can see it happened.
Related: the seed keyword is no longer screened when it is only a concept label — it still is when a listing will put it in your marketplace copy. That removes one billed clearance call per text-only design that was answering a question nobody asked.
A text design printed step timings with no header, so the run log never showed the wording, font, colour, finish or canvas. It now opens with the same block the image path uses and closes with the rendered lines, size, DPI, file size, transparency and a link to the print file. Failures name the stage and confirm nothing was charged.
artOutput: "textOnly" calls no image model and no language model, but the run still demanded the
image-generation and language-model keys before starting — and with transparency: "native" it
demanded a third. Those checks now apply only to the path that spends them.
Every run that produces artwork now writes two CSVs you can download straight from the run:
| File | Use it for |
|---|---|
upload-lazymerch-mba.csv | LazyMerch, in its own documented MBA column order |
upload-generic.csv | Any other uploader, or your own spreadsheet workflow |
The image link goes in the sheet, so there is nothing to download first and no local file paths to paste. LazyMerch's own specification allows a direct, unauthenticated link to the image in place of a path, and the artwork this Actor writes is exactly that.
Columns only you can answer — the product-settings template name, fit type, placement arguments — are left blank rather than guessed. A wrong template name silently falls back to your default and a guessed fit type merchandises on your behalf, so blank is the honest value.
The generic sheet carries the trademark verdict in the same row as the copy it applies to. A
bulk uploader will publish a caution phrase without comment; keeping the verdict beside the
listing is what puts it in front of you first.
Only designs with a validated print file are included — a row pointing at failed artwork would import as a listing with no image. Quoting is RFC 4180, with a UTF-8 BOM and CRLF so Excel opens it cleanly; listing copy routinely contains commas and quotes, and a shifted column would upload the brand into the title field without anything looking wrong.
An audit of every artwork dimension against its publisher found that the Merch presets for Pullover Hoodie, PopSockets Grip and Tote Bag matched no published source at all. They have been removed rather than adjusted, because there is no figure to adjust them to: Amazon publishes its template sizes behind a seller login, and the public numbers contradict one another.
| Preset | We shipped | Source A | Source B |
|---|---|---|---|
| PopSockets | 1000 × 1000 | 1876 × 1876 | 485 × 485 |
| Tote Bag | 4500 × 5400 | 4500 × 4500 | 2925 × 2925 |
| Pullover Hoodie | 4500 × 5400 | 4500 × 5400 (back) | 4500 × 4050 (front) |
Merch matches uploaded pixel dimensions exactly — a 4500 × 5399 file is rejected — so a canvas we could not confirm meant a rejected upload the customer had already paid for. That is precisely the failure this Actor exists to prevent, so an unsourced number is worse here than a missing option. The hoodie is the clearest case: our figure is right for the back and wrong for the front, and one "Pullover Hoodie" option cannot express both.
Amazon Merch now offers standard and premium t-shirt, both 4500 × 5400 — the only Merch canvas
every independent source agrees on and which is proven by real accepted listings. An API caller
passing a withdrawn type falls back to that canvas rather than erroring, and the row's
canvasPreset states which was used.
Redbubble was re-verified against its live published table and is unchanged — all nine product canvases, the 300 MB / 13500 px ceilings and the DPI disclaimer still match verbatim. Redbubble documents everything publicly, which is why it keeps nine canvases where Merch keeps one.
Every shipped canvas now carries a source and a verification date, pinned by a test.
artOutput: "preview" and the design-preview-generated event have been removed. The cheap
preview ran on a fast image model that renders misspelled words in-image — and print-on-demand is
overwhelmingly typographic, so a design whose lettering is wrong is unusable at any price. A tier
that reliably produces unusable output should not exist, let alone bill for itself.
Because the preview was charged before the print stage, a print-ready run always paid for it too. Removing it takes that charge off every artwork run: the art tier now bills exactly once, after the finished file has passed its dimension, transparency and file-size assertion.
artOutput is now none or printReady.error rather than partial, because
there is no half-deliverable left to justify partial. Nothing is charged for it.previewImageUrl has been removed from the dataset schema and the upload manifest.None of this reached a live pricing record — the event was removed before the Actor's first pricing publish, so no existing run or invoice is affected.
gpt-image-1.5The premium transparency: "native" path now generates with gpt-image-1.5. Measured live on the
same prompt and request parameters as the model it replaces:
| $/image | time | semi-transparent edge | |
|---|---|---|---|
gpt-image-1 (was) | $0.2496 | 51s | 15.07% |
gpt-image-1.5 (now) | $0.2111 | 43s | 10.13% |
Cheaper, faster and a cleaner edge at the same time — the semi-transparent band is what prints as a
faint halo on a coloured garment. gpt-image-2 was tested and rejected: it refuses transparent
backgrounds outright, so the newest model is not usable on this path at all.
Price is unchanged. The saving is ours to absorb, not a reason to re-open a pricing record.
The prompt sent to the native generator ended with "on a plain flat white background" and then "fully transparent background" — the keyed path's instruction and the native one, in the same sentence. Each path now gets only its own. Measured honestly: this did not move the halo (8.83% → 8.86%, i.e. no effect), because the halo is glow the model draws into the artwork rather than a background fill. Glow, neon and bloom are now banned in the design brief alongside gradients and soft shadows, which is where that problem actually lives.
Every event price has increased substantially. The Actor now prices against what its output is worth to a seller rather than against what it costs to produce, and the ladder is set so that a complete listing — one niche, one clearance, one listing pack and one print-ready file — clears a fixed revenue floor even on the smallest possible run.
| Event | Was | Now (base rate) |
|---|---|---|
| Actor start | $0.005 | $0.005 |
| Niche demand & competition analysis | $0.030 | $0.65 |
| Trademark clearance screening | $0.020 | $0.35 |
| Platform-compliant listing pack | $0.025 | $1.85 |
| Design preview | $0.035 | $0.75 |
| Print-ready artwork | $0.120 | $5.00 |
| Print-ready artwork, native transparency | $0.550 | $6.75 |
A minimum complete listing moves from $0.20 to $7.86 at the base rate; a typical single-keyword run, which screens about four phrases, from $0.26 to $8.91. That is a 34–39× increase depending on run shape, and it is a repositioning rather than an adjustment: the Actor leaves the cheap-utility band it launched in and becomes a premium tool for sellers who intend to publish what comes out of it. Higher-volume accounts pay less than the base rate — see the volume ladder below.
Nothing about the product changed — no tier was added, removed or degraded, and every quality
decision behind the print tier stands. What you are charged for also has not changed: billing is
still per validated artifact, a failed print file still costs you nothing, and the run modes still
let you buy only the stage you need. research remains the cheap front door at one charged event
per keyword.
The old pricing doctrine exception on the print tier is retired. It existed because $0.120
exceeded the internal $0.025–$0.05 band and had to be justified on cost grounds — the event would
have lost money inside the band. At $5.00 that argument no longer applies and it would be dishonest
to keep making it. The prices are a deliberate value-based repositioning, not a cost-recovery
exception. Full reasoning in .actor/pricing-rationale.md.
Higher-volume plans now pay less per event. Every metered event steps down from the base rate as your Apify plan tier goes up:
| Plan tier | You pay |
|---|---|
| BRONZE (base) | 100% |
| SILVER | 95% |
| GOLD | 90% |
| PLATINUM | 85% |
| DIAMOND | 80% |
So a print-ready file is $5.00 at the base rate and $4.00 on DIAMOND; a clearance screen is $0.35 and $0.28. The per-run Actor-start fee stays at $0.005 on every paid tier — it is a fixed fee, not a metered unit.
This discount is the reason the base rate went up. An earlier revision of this pricing set the same price on every paid tier, because the base cleared the revenue floor by under 0.1% and any discount at all would have pushed high-volume accounts under it. Rather than ship a ladder with no ladder in it, the base rate was raised so that the deepest discount still clears the floor: on DIAMOND, the smallest complete listing grosses $6.29 and stays above the target. Customers on the larger plans end up paying roughly what the old flat rate would have charged them, and get a real volume incentive for the first time.
Previously this Actor only knew Amazon Merch on Demand's rules. Amazon Merch is now the default
platform rather than the whole product, and a new platform input selects whose rules your copy is
written and validated against.
limitsApplied on the row says so. A guessed limit produces
listings rejected at upload, which is the exact failure this Actor exists to prevent, so we
support four platforms correctly rather than eight approximately.The print image model was re-selected by a live bake-off across five real typographic prompts, not by reputation. The previous model was disqualified for two failures that no automated check catches: it returned a photo of a t-shirt on a hanger instead of a design, and it misspelled in-image text. The replacement renders clean standalone graphics with correct spelling, at half the latency of its own higher-tier sibling.
The image model and the print canvas have different aspect ratios, so artwork is now fitted inside
the canvas with transparent padding and never stretched — distorting the lettering is not a trade
worth making on a typographic design. Every print row carries a printGeometry object with the
placed pixel size, the printed size in inches, the padding, and the native sampling density behind
the declared 300 DPI.
New transparency input. keyed (the default) generates and then cuts the background out.
native generates with real alpha from the start, so there is no cutting step and nothing can be
removed wrongly — the cleanest possible edges on a garment. It is billed as its own
design-print-ready-native-alpha event rather than sharing the standard print event.
A live registered mark whose name is canonically identical to the phrase being screened now blocks at every risk threshold. Previously the screen counted live marks, which treated "FELINE FINE" and "FELINE FINE CATTERY SUPPLIES" as the same signal — they are not. Names are canonicalised (case, punctuation and spacing normalised) before comparison, apparel marks decide the verdict when several match, and the whole check degrades to a clean neutral result if the registry is unreachable. It reuses the mark data already fetched, so it costs nothing extra.
permissive still exists to let you proceed past a live mark in an unrelated class. It does not
let you print someone else's registered mark verbatim.
Amazon Merch niches now carry a computed pass/fail alongside the scores: more than 10 reviews as proof somebody actually buys in the niche, then under 1000 competing results, leaders under 100 reviews, and a $16–25 price. It is arithmetic on the live rows, not a model's opinion — the model only explains the verdict. Gated to Amazon Merch, because the price band is an Amazon Merch t-shirt price and borrowing it for a Redbubble scarf would reject good niches for no reason.
The trademark-phrase-cleared cost basis is no longer an estimate. It is $0.0054 per registry
call, so a default screen — the availability lookup plus the active-mark search — costs $0.0108.
That closes the unverified-cost flag raised at launch.
merchLimits is now platformLimits. backendKeywords is now tags — the field print-on-demand
platforms actually have. The key-value store manifest is now upload-manifest.json.
clear, caution or blocked instead of a bare availability yes/no, because a registry can report a phrase "available" while a live pending mark sits against it.research, clearance, listing, design or full, so you pay only for the stage you need.The launch entry carried two pricing notes: a doctrine exception for design-print-ready at
$0.120, and an unverified-cost flag on trademark-phrase-cleared at $0.020. Both are obsolete.
The registry cost was verified at $0.0054/call, and every price was subsequently reset to a revenue
target — see "Repriced to a revenue target" under [Unreleased] for the current ladder. The
launch-era figures are not repeated here so they cannot be mistaken for live prices.
Two build decisions from that entry still stand and are restated so nobody "optimizes" them back:
Failure absorption: if generation succeeds but the file fails its dimension, transparency or size assertion, the customer is charged nothing even though the vendor cost was already incurred. Charging for an unusable file would be the wrong behaviour.