▲ATIYO

Creative operations for DTC brands

Approve Shopify AI Product Scenes Without Overwriting the Source Image

Shopify’s AI media tools make it easy to place an existing product into a new setting, but generation is only the first step. A dependable workflow must preserve the verified source photograph, retain viable candidates before they disappear, document how each derivative was made, and require product, brand, and channel approval before publication.

By ATIYO editorial system Source and product-claim checks completed

Direct answer

The short answer: treat each generated scene as a derivative asset

Do not edit the only copy of a verified product photograph. Lock that photograph as the source, generate scenes from a working copy, and use Shopify’s “Save as new” option for every candidate you may want to review. Shopify warns that editing an existing file can affect everywhere the file is used, while saving as new creates a separate file and preserves the original. It also states that unused generated scenes are unavailable after the editor closes, so retain promising candidates during the generation session rather than assuming you can retrieve them later. For every saved scene, record the source asset, exact prompt, SKU and variant, operator, generation date, intended placement, and candidate number. Compare the derivative side by side with verified photography, checking color, proportions, logos, packaging text, materials, included components, scale, shadows, reflections, and any background objects that could mislead a customer. Move the candidate through explicit states such as GENERATED, PRODUCT_QA, BRAND_REVIEW, CHANNEL_REVIEW, and APPROVED_PDP or APPROVED_PAID. Approval should be channel-specific: an image suitable for organic social may not be appropriate as a product page’s primary image or a shopping-feed image. Keep verified photography available even after an AI lifestyle scene is approved.

01

1. Lock a verified source before opening the AI editor

The original photograph should remain an unchanged reference, not become the first version in a chain of destructive edits.

Start by identifying the photograph that accurately represents the sellable SKU and variant. Give it a stable asset ID and record the product color, material, size, package configuration, and photography date when those details are relevant. A useful filename is SKU123_BLUE_SOURCE_VERIFIED_v001.jpg. The word SOURCE distinguishes evidence of the actual product from generated derivatives; VERIFIED indicates that someone has checked the image against the sellable item.

Keep a controlled copy outside the editing session. Shopify’s file-management documentation says merchants can download an original, unoptimized file. That gives the team a recoverable copy if an operational mistake occurs in Shopify. Do not rely on filename differences alone: mark the source as prohibited from direct editing in your asset register, brief, or project instructions.

This distinction matters because Shopify explains that edits to an existing file can apply everywhere the file is used in the store. Its product-media guidance recommends “Save as new” when the existing file must not be overwritten. Establishing the source-versus-derivative rule before generation is safer than trying to reconstruct lineage after several similar images have accumulated. See the internal guide to ecommerce asset version control for a broader naming and preservation structure.

  1. Assign the source a permanent asset ID.
  2. Verify the SKU, variant, color, packaging, and included components.
  3. Download or retain a controlled copy of the original.
  4. Rename or label it SOURCE_VERIFIED.
  5. State explicitly that the source must not be edited or replaced.

02

2. Generate from a working copy and capture the exact prompt

A generated scene is more useful when the team can identify precisely what produced it and which real product image it was derived from.

Shopify Magic’s file editor can replace a background, generate a setting, modify lighting or style, and extend an image beyond its original borders. Shopify recommends concise natural-language prompts that describe details such as the scene, lighting, colors, framing, and camera angle. Its guidance also suggests supplying real product dimensions when scale matters and explaining how the product should sit in the scene to reduce a floating appearance.

Record the exact prompt rather than a summary written after the fact. Small wording changes can correspond to materially different staging, shadows, scale, or product placement. The record should also include the source asset ID, operator, generation date, Shopify Magic as the tool, candidate number, intended SKU and variant, and proposed destination.

For example: 8 oz candle sitting on a walnut table, warm window light, neutral kitchen background, front shot, professional product photography. Pair that prompt with a candidate identifier such as SKU123_BLUE_AI-KITCHEN_c01_v001. If a later candidate uses a revised prompt, preserve both records instead of replacing the first prompt. This creates a basic provenance trail. The AI creative provenance workflow provides a fuller structure for connecting a generated asset to its source and generation context.

  1. Create a working copy or start from the protected source without saving changes over it.
  2. Write a prompt containing the scene, lighting, framing, angle, and relevant scale instructions.
  3. Copy the exact prompt into the asset record.
  4. Record the source ID, tool, operator, date, SKU, variant, and intended placement.
  5. Give each generated candidate its own sequential identifier.

03

3. Save every viable candidate before closing the editor

Do not use the generation screen itself as an approval queue. Retain each promising candidate as a separate file while the session is still open.

Shopify says its media generator produces one scene at a time and that unused scenes disappear when the editor closes unless they are saved. In Shopify’s unified file picker, a generated file is discarded unless the user explicitly chooses to keep it. If a candidate might be useful, retain it immediately; do not assume that a teammate can return to the same alternatives tomorrow.

Use “Save as new,” not “Save,” whenever a candidate enters review. A filename such as SKU123_BLUE_AI-KITCHEN_c01_v001_REVIEW.jpg communicates the SKU, variant, generated setting, candidate number, version, and current status. If the image has a transparent background, Shopify advises saving it as PNG because other formats do not preserve transparency.

Avoid putting sensitive prompts, internal review comments, or confidential campaign plans into public-facing file fields. Shopify describes product media as files intended for storefront use. Keep internal records in a controlled project or asset system, with only the publishable image moving into customer-accessible locations. ATIYO’s asset Vault documentation is relevant when a team wants to preserve working assets alongside its creative context.

  1. Retain each viable candidate during the active editing session.
  2. Choose “Save as new” rather than replacing the existing file.
  3. Use a unique candidate and version number.
  4. Use PNG when transparency must be preserved.
  5. Store prompts and internal review notes separately from public storefront media.

04

4. Run product-fidelity QA before evaluating creative taste

The first review question is not whether the scene looks attractive. It is whether the depicted product still matches the item a customer will receive.

Compare the generated candidate directly with the verified source at a useful zoom level. Shopify says generated images default to approximately one megapixel based on the source aspect ratio: larger source images are scaled down and smaller ones are scaled up. Reviewers should therefore inspect fine packaging details, label text, stitching, hardware, and surface texture rather than assuming the source image’s detail has carried through unchanged.

Check the silhouette, proportions, color, pattern, material, logo shape, logo placement, package text, warnings, closure, handle, seams, quantity, and contents. Inspect transparent areas, reflections, contact shadows, and the relationship between the product and surrounding objects. A generated hand, table, appliance, or room can make the product appear larger or smaller than it is. Background accessories can also imply that extra items are included.

Look for functional and commercial changes, not just visual artifacts. Reject an image if generation removes a component, invents a feature, changes a variant, alters a safety label, or creates an unsupported health, environmental, or performance implication. Record a specific rejection reason—such as logo misspelled or lid shape differs from source—so the next prompt or edit addresses an observed defect.

  1. Compare the candidate side by side with SOURCE_VERIFIED.
  2. Inspect labels, logos, hardware, materials, and fine details at high zoom.
  3. Verify the correct SKU, variant, quantity, and package contents.
  4. Check scale, contact shadows, reflections, and transparency.
  5. Identify props that might be mistaken for included accessories.
  6. Reject unsupported product features or claims with a specific reason.

05

5. Move the image through explicit review states

Separate product accuracy, brand suitability, and destination requirements so that one reviewer’s approval cannot be mistaken for permission to publish everywhere.

A practical queue is GENERATED, PRODUCT_QA, BRAND_REVIEW, CHANNEL_REVIEW, followed by destination-specific outcomes such as APPROVED_PDP, APPROVED_PAID, or APPROVED_ORGANIC. After use, move the asset to PUBLISHED. Failed or superseded candidates should become REJECTED or ARCHIVED, not simply vanish from the record.

Each state transition should capture the reviewer, date, decision, corrections requested, approved destinations, and any re-review or expiration date. Product QA should be performed by someone able to recognize the actual variant and packaging. Brand review covers composition, styling, tone, and visual consistency. Channel review asks whether the image is appropriate for its specific placement.

Avoid a generic APPROVED status. A stylized scene might be acceptable in an email or social post while being unsuitable as a product page’s primary image. A decision record can say: Approved for Meta prospecting and organic social; not approved for PDP primary image or Google main image. That is more actionable than expecting downstream users to infer restrictions from comments.

  1. Set the initial state to GENERATED.
  2. Require a named reviewer for PRODUCT_QA.
  3. Complete brand review only after product accuracy passes.
  4. Assign destination-specific approval during channel review.
  5. Record publication separately from approval.
  6. Archive superseded candidates while preserving their lineage and decisions.

06

6. Use stricter rules for PDPs and shopping feeds

Keep verified product photography available, and use AI lifestyle scenes as supporting media unless the generated asset accurately satisfies the destination’s requirements.

Google Merchant Center’s image guidance requires the image to display the correct product and variant accurately. Its main-image guidance emphasizes a clear view of the entire product with minimal or no staging, while additional images can present the product in use or within a setting. That makes a generated lifestyle scene a more natural candidate for additional imagery than for the main feed image, although the final decision must reflect the actual image and current destination rules.

On a product detail page, preserve at least one verified photograph that shows the exact item. Do not allow a generated scene to become the only visual evidence of appearance, scale, color, packaging, or included contents. If props could be mistaken as included, use a different image or make the distinction unambiguous in the surrounding product presentation.

Google also instructs merchants submitting generative-AI images to retain metadata identifying the synthetic source, including the appropriate IPTC DigitalSourceType value. Avoid export or optimization steps that strip that metadata when the file will be supplied to Merchant Center. Shopify separately states that it applies an invisible watermark to generated images and that the watermark cannot be removed. Treat platform provenance and your internal generation record as complementary controls rather than substitutes for one another.

  1. Keep verified photography on the PDP.
  2. Prefer exact, minimally staged product imagery for a main shopping-feed image.
  3. Use approved lifestyle scenes as additional media when appropriate.
  4. Confirm that the pictured variant and package configuration match the offer.
  5. Preserve required generative-AI metadata during export and handoff.
  6. Document which destinations have approved each derivative.

07

7. Use a compact approval record that travels with the asset

The record should let a teammate understand the image’s origin, review status, and usage limits without reopening the generation session.

A practical record can read: Asset ID: SKU123_BLUE_AI-KITCHEN_c01_v001; Source: SKU123_BLUE_SOURCE_VERIFIED_v001; Tool: Shopify Magic; Prompt: [exact prompt]; Status: CHANNEL_REVIEW; Product QA: Passed; Brand review: Passed; Approved channels: Meta prospecting, organic social; Not approved for: PDP primary image, Google main image; Reviewer: [name]; Decision date: [date]; Notes: Label verified; background objects are not included with purchase.

Keep corrections attached to the candidate they concern. If the team generates a replacement, create c02 or a new version and connect it to the rejected candidate rather than silently swapping the file. This preserves the reason for iteration and prevents an old approval from being applied to a materially different image.

A spreadsheet, project database, or asset-management process can support this workflow. The essential requirement is consistent identifiers and explicit decisions, not a particular tool. For a small team, one row per candidate is enough if image files are stored safely and status changes have named owners.

  1. Create one record per candidate.
  2. Link every derivative to one verified source asset.
  3. Store the exact prompt and generation context.
  4. Record approvals and restrictions in structured fields.
  5. Create a new candidate or version whenever the image materially changes.

08

8. Where ATIYO fits—and where it does not

ATIYO can organize the creative context around generated scenes, but it does not replace Shopify’s editor or destination-platform reporting.

ATIYO is a creative-strategy operating system for founder-led ecommerce brands, creators, and dropshippers. Teams can organize roadmaps, briefs, brand context, assets, iterations, and reusable learnings in one workflow. For this use case, a source image, generated candidates, prompt records, review decisions, channel restrictions, and iteration notes can be kept together rather than split across chat threads and ambiguous filenames.

ATIYO should not be treated as an automatic Shopify publishing connection. It does not connect to ad accounts, buy media, calculate ROAS, or know campaign results unless a user records them. Media performance remains in the ad platform and ATIYO preserves creative context and learnings. When results matter to the next creative decision, a user can record the relevant observation alongside the asset without implying that ATIYO independently retrieved or verified it.

If your team needs a shared queue for briefs, assets, iterations, and reusable creative learnings, start ATIYO’s card-confirmed seven-day trial before billing begins.

  1. Keep source and derivative assets linked to the same creative initiative.
  2. Attach prompts, review notes, and channel permissions to the relevant candidate.
  3. Record performance observations manually when they inform future iterations.
  4. Continue using Shopify for Shopify media editing and publication.
  5. Continue using each media platform as the source for delivery and performance data.

Frequently asked questions

Questions about this workflow

Should I edit the original Shopify product image directly?

No. Preserve the verified source and use “Save as new” for any generated candidate that may enter review. Shopify warns that edits to an existing file can affect everywhere that file is used.

Can I recover an AI-generated scene after closing Shopify’s editor?

Do not count on it. Shopify says unused generated scenes are unavailable after the editor closes unless they were saved, and generated files in the unified picker must be explicitly kept.

What should an AI product-scene reviewer check?

Compare the candidate with verified photography. Check the variant, color, proportions, materials, logos, text, hardware, package contents, scale, shadows, reflections, transparent areas, and props that might imply additional items are included.

Can an AI-generated lifestyle scene be the primary product image?

It should only be used if it accurately represents the offer and satisfies the destination’s requirements. A safer default is to keep verified, minimally staged photography as the primary evidence of the product and use approved generated scenes as additional media.

Does ATIYO automatically publish images to Shopify or retrieve ad performance?

No. ATIYO organizes creative assets, context, briefs, iterations, and recorded learnings. It does not publish Shopify media automatically, connect to ad accounts, buy media, calculate ROAS, or know performance unless a user records it.

Primary and official sources

Sources used in this guide

External product facts were checked against the organizations’ own documentation. Features can change; confirm current details before making a purchase or campaign decision.

  1. Shopify Help Center — Media generation in the file editor Consulted for generation capabilities, prompt guidance, temporary alternatives, image sizing, transparency, invisible watermarking, and commercial-use information.
  2. Shopify Help Center — Adding product media Consulted for the warning that edits can affect uses of an existing file and for Shopify’s “Save as new” preservation guidance.
  3. Shopify Help Center — Uploading and managing files Consulted for downloading original, unoptimized files.
  4. Shopify Help Center — Product media Consulted for the role and storefront use of product media.
  5. Google Merchant Center Help — AI-generated content Consulted for requirements concerning retained generative-AI metadata and IPTC DigitalSourceType.
  6. Google Merchant Center Help — Image link Consulted for product-image accuracy, correct-variant, main-image, and additional-image guidance.

Move the plan out of scattered sheets

Run the roadmap, briefs, assets, and learnings in ATIYO.

ATIYO keeps the brand context and production decisions connected. It does not buy media, connect to ad accounts, or invent performance results.

Start your 7-day trial → Book an onboarding call