The problem this solves
Cart recovery emails all reach for the same image: the product on white, cropped from the catalog, identical to the photo the shopper already scrolled past on the way to abandoning. If the words of a recovery email must work harder than a product page, the picture should too - yet producing three distinct, intentional images per flow means a shoot or a designer's week, so every email in the sequence recycles the same catalog crop and the visual layer of the flow persuades no one.
This matters because the three emails of a recovery arc are doing three different jobs, and one image cannot serve them all. The first email's gentle urgency, the second's social weight, the third's honest last call - each frames the product differently in words while showing it identically in pixels. The mismatch between escalating copy and static imagery is subtle, and readers feel it: the sequence reads as one email sent three times.
Generated product scenes change the production economics. The product can be placed in a context that serves each email's specific argument - in use, in life, in the moment the shopper imagined when they added it to the cart - at drafting cost rather than shoot cost. Paired with proof mined from real reviews, the flow argues in both channels at once, and every visual passes your review before any shopper sees it.
How the mission runs
- Reviews are mined for the proof layer. Web Research works through your reviews and public mentions to extract the objections shoppers voice and the customer language that answers them. This evidence base shapes both channels of the flow: the copy quotes and paraphrases it, and the scenes are briefed to depict what reviewers actually praise.
- Three emails are drafted with escalating jobs. Email one leans gentle urgency: the cart is waiting, and the top objection gets answered with proof. Email two carries social weight, built from what buyers consistently say. Email three is the honest last chance, factual about expiry or stock. Each email's copy defines what its image must argue.
- A distinct scene is generated per email. Image Generation produces a product scene matched to each email's job: an evocative in-context framing for urgency, a scene that echoes the review language for social proof, a clean and direct treatment for the last chance. Scenes are generated against your brand's look, with alternates where a concept could go two ways.
- You review copy and visuals as pairs. Each email arrives for review with its scene attached, because the unit of judgment is the pairing rather than the parts. You approve, swap an alternate, or redirect a concept, and regenerated scenes return under the same pairing. Every image a shopper could ever see passes this human gate.
- The flow is staged in Klaviyo with visuals linked. On approval, the agent builds the three-email flow in Klaviyo - trigger, delays, copy, and the approved visuals linked in place - and reports exactly what was created. The flow is staged rather than live, so your platform-side walkthrough is the final check before any recovery email sends.
The prompt
This is the exact objective the agent receives. Swap the obvious placeholders for your own domain, segment or channel and run it as-is from the console, Slack, or the API.
What comes back
A cart recovery flow staged in Klaviyo where words and images escalate together: three emails built on review-mined proof, each paired with a generated product scene designed for that email's specific job - urgency, social proof, last chance - every pairing human-approved before the build. The review evidence base and the approved scene files remain yours for reuse across ads and product pages.
Make it yours
- Generate scenes per hero category: your two best-selling categories get their own scene sets, so the visual context matches what is actually in the cart.
- Test the visual layer itself: run the same copy with generated scenes against catalog crops for a period, and let the recovery numbers judge the production change.
- Extend the look outward: once scenes are approved, have matching variants generated for the retargeting ads that chase the same abandoned carts.
Frequently asked questions
Will generated scenes misrepresent my product?
The scenes contextualize your product rather than redesign it, and fidelity is an explicit review criterion: any render that distorts what the product looks like gets rejected or regenerated at the pairing review. Nothing ships without your approval, so the accuracy gate is a human looking at every image.
Do I need to disclose that the images are generated?
Norms and rules on disclosure vary by market and are evolving, so the playbook stages everything for your review and leaves the disclosure decision with you and your counsel. Because scenes are contextual settings for a real product, many teams treat them like staged photography, but that judgment is yours.
What if none of the generated scenes fit my brand?
Scene generation is briefed from your brand's visual language, and the pairing review exists for exactly this: redirect the concept in plain words and regenerate, or approve copy-only and add photography later. The flow's build waits on your approval, so an unsatisfying image can never force its way into a send.