Product Image Compliance Editor.

End of run

A product image that actually passes the retailer’s intake review: correct background (pure white or generated lifestyle scene), correct dimensions, correct format, free of low-resolution flags and demographic-rule violations, with the original product untouched in the center of the frame. Ready to upload to the retailer’s catalog without bouncing.

~4 minutes of compute per image, $0.44, three models cooperating (Claude for analysis and compliance, Gemini 3 Pro Image for the background and lifestyle generation).

Challenges it solves

Every retailer wants something different. White background at a specific pixel ratio. Lifestyle scene with no people of a specific demographic. Specific aspect ratio. Specific format. The agent collects requirements as input and routes through a Validation Router that catches contradictions, conflicting requirements, and demographic-rule violations before generation even starts, so you don’t pay for a render that was always going to fail QA.

Source images are not retailer-grade. Vendor product photos arrive in mixed resolutions, on messy backgrounds, in the wrong aspect. The agent runs a Resolution Check, refreshes the product, removes the background, re-renders against a clean or lifestyle background, then composites at the required dimensions.

The product can’t drift. The single non-negotiable: the physical product in the final image must match the source. The refresh / re-upload / background-render loop is designed so the generated background is layered around an unmodified product cut-out, not regenerated from scratch.

Compliance is verified, not assumed. Before returning, the agent runs a Compliance Checklist against the original requirements (dimensions, format, background type, content rules), then routes to Review & Accept or Revised Requirements for another pass.

Under the hood

One ambitious workflow, deeply branched. Linear at the start (Collect Inputs → Analyze Product → Get Dimensions → Validation Router), with error branches for each kind of bad-input condition. The image-modification arm runs Remove Background → Background Render → Composite & Render → Apply Effects → Convert Format, with a Generate Lifestyle BG branch for retailers that want scene-based imagery. Final loop: Compliance Checklist → Display Results → Review & Accept, with Revised Requirements feeding back in if the checklist fails.