Reviewed 9 September 2026. This is an editorial workflow based on Google and IPTC documentation, with proposed studio checks. It reports no client campaign, Merchant Center approval, or performance lift.

Start with the product, then choose the image slot

A studio brief should name the exact product variant and the destination field before production begins. Treat an approved product photograph or verified product reference as the comparison source. A persuasive scene cannot resolve uncertainty about what is actually being sold.

Google’s main-image guidance requires an accurate product depiction and prohibits promotional overlays, watermarks, and borders, subject to its documented exceptions. Google addresses AI disclosure labels separately. Check the current category-specific instructions rather than treating a social campaign crop as a feed-ready master. Google: image link.

Additional images have a different job. Google permits product staging, use scenes, and detail views through additional_image_link, with up to ten additional images. Those exceptions do not remove the other applicable image requirements. Google: additional image link.

Our studio recommendation: deliver a clear main product view and separate context/detail variants. Label the destination in the handoff record so a feed operator does not accidentally replace a product master with a campaign composite. This is a proposed production convention, not a requirement to use a particular photography method.

Accept, repair, or reshoot

Use these decisions for each final asset. They are editorial acceptance criteria, not Google’s automated approval rules.

Accept for the next check: the product owner can verify the shape, color variant, material, printed details, and included accessories against the reference. Any scene or pose supports the product without obscuring a material feature. Acceptance here means continue to file and feed checks; it does not mean approved by Google.

Repair and review again: the product is demonstrably correct, but the crop, background, contrast, or export needs work. Keep the approved product reference beside the repair. Recheck the whole image after editing; do not assume a local correction leaves every other feature intact.

Reshoot or rebuild from a verified reference: the image invents a seam, changes a closure, distorts packaging text, substitutes a color, or shows an accessory that is not included. If the team cannot establish what a detail should look like, generating another plausible version does not establish accuracy. Obtain a better reference or photograph the actual item.

For example, a fictional jacket concept with an extra pocket fails the product check even if its lighting is excellent. A correct jacket cropped too tightly may be repairable. These are hypothetical review cases, not examples from a measured 404 Models campaign.

For testing the same synthetic character across scenes, use our separate character consistency protocol. This guide concerns one product image’s handoff into Google Shopping.

Preserve AI-origin metadata through the handoff

Google requires metadata identifying images made with generative AI and says not to remove embedded digital-source tags. Its guidance names IPTC Digital Source Type and distinguishes generated, composite, and algorithmic origins. A caption or filename is not the embedded metadata Google describes. Google: AI-generated content.

Have the production operator document how the asset was created, then verify that the metadata truthfully describes that process. Do not apply the same origin value blindly to every photograph, retouch, and composite. The IPTC Digital Source Type vocabulary defines the terms; the site’s capitalization in prose is not an instruction to invent a new metadata value.

Check the final delivery path, not just the master in the studio folder. Our recommended test is to export the approved asset, pass it through the actual hosting or image-conversion workflow, retrieve the exact file intended for the feed, and inspect its embedded source metadata. Record whether it survived. If it did not, repair the export or hosting configuration and repeat the check before handoff.

A valid origin tag does not establish product accuracy. Likewise, a correct-looking image does not establish that its provenance survived conversion. Keep the visual and metadata decisions separate. For broader signed-provenance workflows, see our C2PA field guide.

Check the file Google will fetch

Ask the feed operator to verify the actual hosted asset without an authenticated studio session. Confirm that the response is an image, that the format is supported, and that the image crawler is not blocked. Google’s image-link specification also covers dimensions, file size, and image URL handling. Google: image link.

There is a dated size transition to plan for. Google permits the existing minimums of 100 × 100 pixels for non-clothing images and 250 × 250 for clothing until 31 January 2027, when the 500 × 500 minimum begins for all products. YouTube Shopping ads on TVs already require at least 500 × 500. Check the current requirements at release time; avoid enlarging a small thumbnail simply to meet a number. Google: minimum image size transition.

After submission, the merchant’s authorized operator should inspect the actual item status and any image issues in Merchant Center. A successful file download or a studio sign-off cannot substitute for that account-level result. This article does not inspect or certify any merchant account.

A compact studio handoff record

For each asset, keep a private operational record containing:

• The product variant, approved reference, and intended main or additional image field.

• The creation method, material edits, and final asset version.

• Product-owner decision: accept, repair, or reshoot, with the specific reason.

• File-format, dimension, and public-fetch checks on the delivered file.

• Embedded AI-origin metadata check after the final transformation.

• Named studio reviewer and feed operator, review date, and unresolved issues.

• Merchant Center status once the authorized operator checks it.

Keep customer records and unpublished product material out of public analytics. For portfolio reporting, use aggregate asset counts and issue categories; do not publish raw handoff records.

What to do next

If you are still deciding where synthetic imagery belongs, start with AI models for ecommerce product photography. For a studio discussion, review 404 Models’ ecommerce approach and bring an approved product reference plus the intended image placements. No approval, campaign return, or Google visibility outcome is promised.