Every image in a bulk run is scored against the garment you uploaded. The shots that miss the bar are flagged; the shots that pass are ready to download. Nobody has to open twenty cards to find the two that went wrong.
The moment bulk generation works, the work moves. You are no longer waiting on images; you are sitting in front of a grid of them deciding which are good enough to list. On a twenty-garment run that is twenty judgements. On a hundred-SKU catalogue it is the whole afternoon you thought you had saved.
And it is judgement of exactly the kind people are bad at: comparing a generated image against a flat garment photo, at speed, for the fourth hour running, looking for a print that has shifted scale or a colour that has drifted half a shade. The misses that matter are the subtle ones, and subtle is what gets missed at speed.
So the useful thing is not another quality claim. It is to be told which two of the twenty need a human, and to be handed the other eighteen finished.
The review bar appears above your results, and only when something is actually flagged. A clean run does not grow UI.
One toggle hides everything that passed. On a fifty-shot run you are looking at the three that need you, in the same grid, in their original positions.
Re-run only the flagged garments. Each card stays where it was, so retrying three does not reshuffle the other forty-seven and lose your place.
The images that cleared the bar download immediately as a set, without waiting for the retries to finish. The good work does not queue behind the bad.
The check compares the generated image with the garment photo you uploaded — colour, print and construction — rather than asking whether the picture is pretty. Fidelity is the thing that causes returns, so fidelity is the thing measured.
The bar reads "3 of 20 need a look — 17 passed and are ready to download". That is a decision you can make in a second, and a number you can report.
Retrying a subset re-runs in place. The most annoying failure mode of batch tooling is a re-run that reorders the grid and makes you re-find everything.
Every output is named from the file you uploaded it as, through the retry and into the download, so a batch arrives already matched to your SKU codes.
Download a batch and the accompanying manifest.csv carries the source file, the output file, the quality verdict and the generation id — so your own tooling can act on the QA result, not just your eyes.
The scoring and the review bar are part of bulk generation, which is on every plan including free. What Pro and above add is the automatic free re-roll of a below-bar image before you ever see it.
| Fidelity scoring & batch review | Quality mode | |
|---|---|---|
| What it does | Scores every result against your garment and flags the misses so you can filter, retry or ship | Re-generates a below-bar image once automatically and keeps the better attempt |
| Plans | Every plan, including free | Pro and above |
| Extra credit cost | None | None — the retry is free |
| When it runs | After the batch, in your hands | During the generation, before you see it |
| Daily limit | None | 3 reviewed retries a day on Pro, 5 on Business and above |
Full detail on the automatic layer is on the quality mode page.
The fidelity score and the review bar are part of bulk generation, which is available on every plan including free. There is no extra credit cost for being scored. What Pro and above add is quality mode, which re-generates a below-bar image once automatically before you ever see it.
Your own garment photo. The check compares the generated result with the garment you uploaded on colour, print and construction, rather than judging whether the image looks good in the abstract. Fidelity to the product is what causes or prevents a return, so that is what gets measured.
Only those garments are re-run, and each card stays at its original position in the grid. Retrying three out of fifty does not reshuffle the other forty-seven. The retries are normal generations and cost the normal credits.
Yes, and that is the point of separating them. The images that passed download as a set immediately, so the work that is finished is not held up by the work that is not.
Yes. A batch download includes a manifest.csv listing the source file, the output file, the quality verdict and the generation id for every image, so your own tooling can route the failures without a human re-reading the grid.
No. It means the automated check found a measurable difference from your garment and wants a human to look. Some flagged shots are fine and some clean ones will still not suit you — the bar is there to direct attention, not to make the final call.
No. If every shot passes, no bar appears at all. A clean run should not grow extra interface for you to dismiss.
Yes. Results can be filed into a named project and a whole project exports as one ZIP of up to 50 images with its manifest, so a reviewed batch goes straight into the bucket for that drop.
Five free credits on signup. Send through a handful of real SKUs, not a test image.