Shopify product photo workbench with AI scene options

I want an internal product photo workbench for my Shopify store. A merchandiser should be able to work through the catalog from one screen instead of opening products one at a time. The layout is two panes: a product list on the left, and a photo workbench for the selected product on the right.

The left pane lists products from Shopify using List Products, with filters for vendor, product type and status. Add a setting for the minimum number of images a product should have, defaulting to 3, and flag every product that falls below it with a clear badge showing its current image count. The list also shows each product's review state (not reviewed, reviewed, or skipped) and can be filtered by it, so a session can start with only the products nobody has looked at yet. Show a progress summary for the current filter set, such as 24 of 180 reviewed, so the team can see how far through the catalog they are.

Selecting a product loads its current photos and variants with Get Product and shows them in the workbench, along with the title, vendor, type and status. The merchandiser picks one existing photo as the packshot to work from.

From that packshot I want to generate several competing scene options in one go using Black Forest Labs Generate or Edit Image FLUX.2 [pro] in edit mode, passing the existing photo as the reference image so the product itself stays consistent and only the scene around it changes. Provide a few saved scene presets (clean studio white, kitchen counter, outdoor daylight, styled flat lay) plus a free-text description, and let the merchandiser choose how many options to generate, defaulting to three.

The workbench also needs touch-up and reframing tools that operate on a selected variant. Let the merchandiser brush a mask over a distracting background detail and remove it with Erase Object from Image FLUX Tools, or mask an area and describe a replacement with Inpaint Image FLUX.1 Fill [pro]. For channel sizing, use Expand Image FLUX.1 Expand [pro] to reframe a chosen shot to the aspect ratios each sales channel needs, with configurable presets such as 1:1 for the storefront grid, 4:5 for social, and 16:9 for banners.

Black Forest Labs generation is asynchronous: each of those operations submits a task and then polls for the result, so show each variant tile in a pending state and fill it in as results arrive rather than blocking the whole screen. The signed result URLs expire roughly ten minutes after generation, so download the image bytes as soon as a task is ready and store them with the app instead of holding on to the URL.

Generated variants sit side by side with the original in a comparison grid, rendered large enough to actually judge, with a click to enlarge view. Nothing touches the storefront until the merchandiser explicitly approves one. On approve, write the chosen image back to the product with Shopify Update Product, adding it to the product's images rather than deleting the originals, with an option to make it the primary image. At the same time archive a full-resolution copy to Google Drive with Upload File (Multipart), into a folder created per store or per product with Create Folder, naming the file with the product handle and the date. The multipart upload path is limited to about 5MB per file, so keep the archived copy in a compressed format such as JPEG or WebP at a sensible resolution.

Persist review state so the catalog can be worked in sessions. Every product should record whether it was reviewed or skipped, who made the call and when, which variant was approved, and links to the Shopify image and the Drive archive copy. Skipping should be one click with an optional short reason. When someone opens the app again it should come back to the same filters and jump to the first product that has not been reviewed, so a team can divide the catalog and pick up where they left off without redoing each other's work.