How it works

Four steps for you. Fifteen states for the pipeline.

You do the two things only you can do — supply the product and make the final call. Everything between those is tracked, scored and recoverable.

Start free trialSee all features

Your part

What you actually do

Roughly seven minutes of human attention per product.

  1. ~2 min
    01

    Upload assets

    Provide details manually or via our Bulk CSV template (50+ products at once). You can choose exactly how many images to generate (1, 3, 5, or 7) per product to save costs. Reference photos lock on upload — they stay the ground truth for everything generated afterwards.

  2. Automatic
    02

    Research runs

    The pipeline scrapes each competitor listing for imagery, price, rating and review depth, and identifies which visual arguments the category is making that your listing currently is not.

  3. Automatic
    03

    Blueprint and generation

    A seven-slot art-direction plan is written for your specific product, then each slot is rendered against its own prompt. The prompt and the competitive rationale behind it are stored with the image.

  4. ~5 min
    04

    Review and download

    Assets arrive already scored. Approve the set in bulk, or reject any single image with a note that becomes part of the next generation prompt. Then download and list.

Under the hood

The pipeline, in full

Five phases covering fifteen tracked states. Every product moves through the same machine, and you can see exactly where it is.

  1. 01

    Intake

    Raw photos and competitor URLs land in an immutable record.

    • Created
    • Uploading
    • Uploaded
  2. 02

    Research

    Competitor listings are scraped for imagery, price and ratings.

    • Research queued
    • Researching
    • Research complete
  3. 03

    Blueprint

    A 7-slot art-direction plan is written per product.

    • Blueprint created
  4. 04

    Generation

    Each slot is rendered against its own prompt and rationale.

    • Generation queued
    • Generating
    • Generation complete
  5. 05

    QA & review

    Every asset is scored, then approved or sent back to regenerate.

    • Qa queued
    • Qa running
    • Pending human review
    • Approved
    • Finalized

Each state is a real checkpoint, not a progress-bar animation. A run that fails during generation stops at that state, keeps everything already produced, and resumes from there.

The output

What lands at the end

Seven positions, each answering a specific buyer objection.

  1. #1

    Main image

    Product only, pure white background — Amazon's hard requirement for the search thumbnail.

  2. #2

    Lifestyle

    The product in real use, so the buyer can place it in their own life.

  3. #3

    Dimensions

    Scale and measurements called out, which removes the top pre-purchase doubt.

  4. #4

    Features

    Annotated callouts mapping each feature to the benefit it delivers.

  5. #5

    Ingredients / materials

    What it's made of, formatted for the categories where buyers check first.

  6. #6

    Comparison

    Your product against the category alternatives the research surfaced.

  7. #7

    Brand story

    The closing frame — who you are and why this product exists.

The loop

Three ways a review can go

Automation handles the volume. You keep the judgement.

Approve

Accept the asset as final. When every image in the set has passed, one action approves the whole listing.

Regenerate with feedback

Write what is wrong. That note is injected into the prompt for the next attempt, so the retry is directed rather than random.

Resume after a failure

If a stage errors, the run stops in an explicit ERROR state and resumes from that point — it does not restart from the beginning.

Run your first pipeline

Bring one product and the competitors you are losing to. The rest is automatic.

Start free trialSee pricing