When the print doesn’t match the Promise or its Provenance

A customer approves a preview on her phone. A lab prints and embellishes the file it receives. Both did their jobs. When the book disappoints her, neither can show what she was told to expect, what the lab committed to deliver, or whether the cover met it.

A customer builds a photo book on her phone. The app lays it out and auto-enhances her photos. She picks a metallic accent for the matte cover, approves the preview, and orders. The book arrives. Skin tones in the beach photos run warmer than they did on her screen. The metallic accent is flatter than she expected. The lab checks the job against its tolerances and finds it in spec. She still says it doesn’t look like what she approved. The order is invented. The gap is not.

Nobody in that chain did anything wrong, which is the problem. The platform showed a preview on a screen. The lab printed and embellished the file it was sent, inside its tolerances. The customer approved what her screen displayed. What she approved, what the producer committed to deliver, and the evidence the product met that commitment were never joined in one record. Nobody can say where it broke.

Screen-to-print differences are old news in photofinishing. A display emits light; a print reflects it. What’s new is how many decisions software makes before the customer sees a preview. Shutterfly’s Auto-Fill “uses AI to arrange your photos into layouts” (Shutterfly). Its Auto Enhance, which the company doesn’t call AI, is on by default and “automatically applied to all Shutterfly prints and projects” (Shutterfly). When software picks the layouts and adjusts exposure and color, the customer’s approval covers decisions she never made.

Where provenance takes hold

I raised this gap at a provenance roundtable at the Digital Media Leadership Alliance (DMLA) Leadership + AI Summit in New York on September 30 (DMLA). My example was a chain of custody in which everybody handling an image does their job right, but nothing carries the provenance or metadata forward at the handoffs. The result can be dead wrong in the eyes of the consumer while everyone along the chain did their job. DMLA’s members include picture agencies, archives, publishers and creators (DMLA). For them provenance means who owns the image, what it’s licensed for, whether royalties get paid, and whether AI can train on it without permission or payment.

Provenance takes hold where money changes hands. When I asked the table where provenance supports a consequential decision, Mounir Ibrahim of Truepic, sitting beside me, said his company found traction “in decisions of consequence, primarily in financial transactions, where money was being exchanged based on pixels and the data that backed it.” Banks and insurers needed to know an asset was where it was claimed to be, on that date. He has made the same case in print (Tech Policy Press). When Sherif Hanna of Google delivered the table’s report-out to the room, the first barrier he named was “the cost of implementation”: “it takes a lot of work to modify all these applications and systems to understand these provenance signals and carry them forward.” The second was incentives. Why would a vendor put engineers on it, he said, if nobody’s asking for it? A photo product order is money changing hands based on pixels. A remake is the producer paying twice for the same pixels.

What a C2PA manifest proves

Someone at the table said the container already exists: A C2PA manifest. That’s half right. Content Credentials, from the Coalition for Content Provenance and Authenticity (C2PA), are a signed, tamper-evident record of assertions about a file: who captured it, which tool edited it, what was done. But a manifest doesn’t hold the pixels. It holds claims about the pixels. A valid signature proves the record is intact and traceable to its signer. C2PA’s own explainer says provenance “alone cannot tell you whether the digital content is true, accurate or factual” (C2PA). It doesn’t prove the signer owned the copyright or could license it. And it says nothing about how the file will print on foil, canvas or coated board.

A licensed image printed with the wrong color conversion or on the wrong surface effects misrepresents the photographer’s work even when every credit and royalty is in order. Her beach photos need no license. A stock image on a cover does, and the license may set conditions on how it’s reproduced. Either way, the rights record stops at the file. Following images into production means carrying a verifiable record past that point, so an authorized party can see what the customer was shown and approved, what the producer committed to, what was made, and whether it conformed.

Stickers on screen, foil on press

In July 2026, Shutterfly invested in four Scodix Ultra 2500 SHD digital embellishment presses (Scodix). Scodix documents a web-to-print flow in which its effects are “previewed instantly in-browser for customer approval” and “the approved design is sent to production, where Scodix technology applies the effects automatically” (Scodix).

That describes a handoff, not a record. “Embellishment” in Shutterfly’s editor means “stickers or frames that enhance pages”; Shutterfly sells some of them as paid “Metallic Cover Accents” on matte hard covers, and doesn’t say how they’re printed (Shutterfly). In the pressroom, embellishment means foil and coatings. Scodix’s flow is generic. Nothing I found shows how Shutterfly’s editor choices map to what its Scodix presses apply, or one record connecting the layout, the approval, the production file, the embellishment layers and a measured result.

What the remake numbers don’t show

If the foil or the color misses, where does it show up in the numbers? Nowhere the public can see. Shutterfly’s 10-Ks for fiscal 2006 through 2017 put returns, for reprint or refund, at about 1% of net revenues, “less than 1.3%” in 2015 (SEC 2006, 2015, 2017). That covers every product and every reason; it never separates color, cropping, substrate, embellishment, damage or lateness. The company’s last 10-K before Apollo took it private in 2019 drops the percentage (SEC 2018; The Dead Pixels Society).

Mixbook’s guarantee excludes “differences between colors on your screen and in print” and treats slight variation between prints of the same product as normal. Its reprint, credit or refund covers printing, binding and manufacturing defects and shipping damage; for a color complaint, it offers to help (Mixbook). The policy doesn’t say how such complaints are classified inside the company, and I found nothing public that does. Labs may well track them. The published language treats a complaint that the color is off as expected variance, outside the guarantee.

The best appearance-specific signal I’ve found is vendor-reported but still anecdotal. On The Dead Pixels Society podcast, Viesus CEO Servi Pieters described one company whose complaint rates jumped when Viesus automatic image enhancement was inadvertently disabled, then dropped 58% when it was turned back on (Dead Pixels Society). That’s one anonymized account, not an audit. But it connects a specific production step to a customer outcome, which is more than returns data that lumps every cause together can do.

Approval, commitment, proof

The customer approves a preview on a screen, under whatever light she happens to be sitting in, along with a product description and whatever the platform tells her about screen and print. That is a record of what she was shown and told. It isn’t acceptance of a physical appearance target, and it doesn’t capture what she understood. The producer’s commitment is a separate record: the substrate, colorants, embellishment, rendering intent and tolerances it undertakes to hit for the product it described to her. Conformance is the proof: measurement and inspection showing the cover the lab shipped met that commitment. An approval, a commitment, and evidence the commitment was kept.

In a dispute like hers, what’s left is the customer’s memory of a screen against the lab’s print. Nothing I found in the public record shows all three carried through the handoffs together. Neither side can show whether the print missed the commitment, or the commitment never promised what the customer thought she saw.

A pilot small enough to run

I’m not proposing a new standard. C2PA already lets a signer include a hashed reference to an outside record, such as a measurement report, so a later change to it shows. The catch: standard validators don’t fetch or check outside records, so the pilot has to (C2PA). The Creator Assertions Working Group’s draft endorsement assertion records approval of three actions, publishing, transcoding and repackaging, so as written it can’t carry a customer’s order approval (CAWG). The pilot needs its own versioned order-approval record, linked by hash to the production and appearance records. That is a profile built on existing mechanisms, not a new foundation. The open question is whether these mechanisms hold a chain together through a real production run.

So test it on one product: A traceable embellished cover. One licensed image, so the rights record is tested alongside the production record. One customer-approved photo book cover with a CMYK print and one embellishment effect. One lab. One physical reference cover, approved by the lab as its production standard and tied to the product it offered the customer: a defined substrate, stated color reproduction conditions, the one embellishment, measured color and gloss, measured embellishment position, the measurement and viewing conditions, and tolerances, all in signed records. A numbered run of 10 to 25 covers, each measured or inspected against that reference, with the results tied to the cover’s number.

Measuring color takes a spectrophotometer, gloss a glossmeter, and a shifted foil layer a dimensional check or machine vision. No single instrument catches every finishing defect. In this design, each handoff leaves a signed record that includes a hash of the record it builds on, so a change upstream shows downstream: the license; the approved preview, with the product description, the finish selected, any screen-to-print disclaimer the customer was shown, and the approval event itself; the production file, with each conversion from the approved preview recorded; the embellishment layer; the measurement reports; the shipment.

Then break the chain on purpose. Swap the image after approval. Alter the production file. Edit a measurement after signing. Shift the foil on one manufactured cover. The first three are digital, and the signatures should catch them. The fourth is physical, and only inspection can. The pilot succeeds if each planted failure is caught by the control built to catch it, every shipped cover maps to the approved version, and an auditor can reconstruct a dispute without seeing the customer’s name or the lab’s recipe. A pass would show the chain held once, under test, not that it holds at production scale. The pilot also has to report what it cost per order, because that number decides whether anyone does it twice. None of this has been run, and none of it promises fewer remakes.

The pilot repeats one image so the chain can be tested. Real photo book production can’t make a reference cover for every customer’s beach photos. There, conformance would rest on a characterized process, control patches on the sheet, job records, inspection data and stated tolerances. That is a different kind of evidence, and it still says nothing about whether she likes the book. Nor can a signature or an instrument reading tell anyone whether a customer is satisfied, what she perceived, or whether the product description was fair. What the record can do is sort the complaint: the print missed the commitment, or the commitment wasn’t what she was shown. Either way she may still be owed a remedy, and either way the lab learns something it can act on.

Who moves first

For photofinishers and photo book producers, the prize would be the ability to show whether they met the appearance requirements they documented, and a remake rate they can sort by cause. Whether customers pay more for that is for the pilot to measure. For platforms, better evidence of why a complaint happened and what response fits it. For licensors, evidence the image was reproduced as the license and the approved specification required. Those are hypotheses. Testing them takes a platform willing to open one order path, a lab willing to measure, an agency willing to license one image, and the vendors whose presses and software would carry the records. None has signed up. The physical cost of finding out is one cover and a few dozen prints. The real cost is the integration, measurement, authorization and staff time around them, which is why the pilot has to count it.

If you run a lab, a picture agency, a photo product platform or an embellishment press line and have want to run it or have questions, write to me: carli@nimahunter.com. I look forward to hearing from you.

By Don Carli, Guest Contributor, President & Co-founder of Nima Hunter, Inc.