Every job starts with a decision nobody checks: whether the images you just uploaded can actually produce the thing you have in mind. Most of the time they can. When they cannot, you find out at the end, after the wait.

There are two ways that goes wrong, and the cheaper one is the failure. A Model Swap on a flat-lay has no model to replace, so it fails, and failed images are never charged: you lose the wait, not the credits. The expensive one is the job that works. Two different t-shirts filed as one outfit, a garment folded so its cut cannot be read, a filter that shifted the exact colour you were trying to reproduce. That job completes, bills, and hands you something you cannot use, so you change one thing and run it again.

A readiness check now runs on every job, across all five tools. Press Continue on the upload step and On-Model reads your images, tells you what will not work for the tool you picked, and what to do about it. It takes a couple of seconds, costs nothing on any plan, and never stops you from continuing.

What happens when you press Continue

Nothing waits. Continue advances exactly as it always did, and the check runs behind it while you choose an identity or write your instructions. If it finds nothing worth saying, you will never know it ran.

The On-Model upload step with a panel reading Checking your images, a spinner, a progress bar and two placeholder rows, above the numbered input thumbnails — A couple of seconds, behind whatever you do next.

The advice itself is not new. The Photo guidelines panel in the right rail has always carried four "Works best" and four "Avoid" examples for every tool, photographed rather than described. The trouble with a panel is that reading it is optional and it cannot see what you uploaded. This is the same guidance, arriving at the moment it applies, about your actual images.

Three kinds of message

Everything it tells you has three parts: what it saw, why that matters for the tool you picked, and what to do instead. How serious it is stays our decision, not the model's. The model reports what is in the picture; On-Model decides whether that is fatal, worth a look, or merely worth knowing.

Things that will fail. Held back for inputs that cannot produce output at all: a Model Swap with nobody in the frame, a mannequin where a model should be, two people when only one can be replaced, or a garment you named for recolouring that is missing from some of the images. These come up again at submit, where On-Model asks whether you are sure and reminds you that failed images are not charged, so it costs you credits only if it works.

Things worth a look. The large middle, and where most of the value is. Two different tops in one outfit. A garment running off the edge of the frame. A product photographed against a scene busy enough to compete with it. None of these stop the job, and the wording says so: "You can continue anyway. This is a heads-up, not a block."

Things that are just information. Upload a top to Flat-to-Model and leave the rest empty, and it will note that the outfit has no bottom and no footwear. Not a complaint: it is telling you what happens next, which is that we style whatever you did not upload and keep it consistent across the whole set. Nothing to fix, and it says so.

The readiness check reading 3 things to look at, listing two warnings and one note, each with what it saw, why it matters and how to fix it, above a numbered strip of input thumbnails ringed in amber — Two warnings and a note on one set. The images a message refers to are ringed and numbered, so you are not left counting.

Because the inputs are numbered and the flagged ones are ringed, a message about images 1 and 2 points at images 1 and 2.

What it catches, tool by tool

Each tool fails in its own ways, so each one is checked against its own list.

Tool	It speaks up when
Model Swap	There is no person in the photo, the garment is on a mannequin, or more than one person is in frame so the choice is ambiguous. It also flags a set showing different people, because one identity goes into every image.
Flat-to-Model	Two different tops are filed as one outfit, a garment runs off the edge, it is folded rather than laid flat, or a heavy filter has shifted its colour.
Create Packshot	Two products share one shot, the product is cropped, or props and background clutter compete with it.
Garment Recolor	The garment you named is not in every image, is partly hidden, is too small in frame to recolour cleanly, or appears twice so the target is ambiguous.
Detail Repair	The detail is too small or too blurry to rebuild, it sits on a crease, or you have marked several problems at once when the tool repairs one region at a time.

A few things are checked everywhere: inputs small enough to limit the detail the output can carry, and sets shot under mixed lighting, which makes a consistent batch harder.

Sometimes the answer is a different tool

Not every flagged image is a bad image. A flat-lay or a ghost mannequin packshot is a perfectly good photograph that happens to be at the wrong door. Model Swap replaces a model, and there is no model in either of them, but Flat-to-Model exists precisely to put that garment on someone.

When the check can tell which tool you actually wanted, it says so and offers the switch, so the recovery is one click rather than a fresh start.

The readiness check on a Model Swap job, reporting in red that the image shows a garment on a ghost mannequin with no person to swap, and offering a button to try Flat-to-Model instead — A packshot in Model Swap is not a bad photo. It is the wrong tool, and the check names the right one.

It never blocks, and it tries to stay quiet

The check is advisory in every case. Continue always continues. Even the problems expected to fail outright only get a confirmation, because you know things about your images that we do not, and the credits are yours to spend.

It is also deliberately conservative. A checker that cries wolf is worse than no checker: the second time someone dismisses a warning that turned out to be wrong, they stop reading warnings at all. So it stays silent unless it is confident, which means it will occasionally miss something it could have caught. We would rather it under-report than teach you to ignore it.

The readiness check only reads your images. It never edits them, never reorders them, never drops one from a job, and never changes a setting on your behalf. Every decision stays yours.

Your library gets a little better each time

To judge a set, the check first has to understand each image in it. That understanding is worth keeping, so it is written back onto the asset and never computed twice.

Open any image you have run through a job and you will find a one-line Description of what is in it, plus a Detected by AI row of the garments, the people and the keywords found in it. That text is also what library search matches on, so searching for "navy knit dress" starts to find the navy knit dress instead of a filename.

An asset information panel in On-Model showing a one-line Description of a grey hooded sweatshirt and a Detected by AI row of garment and keyword badges — Written once, kept forever, and searchable from then on.

These are detected automatically and describe what is in an image. They are a different thing from Tags, which you write yourself to record what you decided about an image. The two sit side by side, and you can filter on both. There is more on the detected side in Your Asset Library, Now Smarter.

Common questions

Does the readiness check cost credits? No. It is free on every plan, including the free one, and it never touches your credit balance. Checking a set you have already checked costs nothing and comes back immediately.

Can it stop my job? No. It never blocks Continue and it never blocks submit. Problems expected to fail outright get a confirmation dialog, and "Submit anyway" is always available.

Does it slow the job down? No. Continue advances immediately and the check runs behind it, so in practice it finishes while you are choosing an identity or writing instructions.

What if it is wrong about my images? Continue anyway. It is tuned to stay quiet unless confident, so a false alarm should be rare, but when one happens it costs you a click. Nothing it says changes how your job is processed.

Which tools have it? All five: Model Swap, Flat-to-Model, Create Packshot, Garment Recolor and Detail Repair. Each is checked against its own list of failure modes.

Does it change my images? No. It reads them and writes a description onto the asset record. The image files themselves are untouched.

Try it now

Open On-Model, start any job, and upload your images as usual. Press Continue and watch the upload step for a moment: if something in the set does not suit the tool, you will hear about it before you spend anything, and if it does, you will not hear anything at all.

New to On-Model? Start with the Flat-to-Model guide, the Create-Packshot guide, or the Model-Swap guide.