From a photo, the AI sees what a trained eye sees at a glance: the garment type, its silhouette, the visible details such as neckline, sleeve shape, length, closure and pockets, and the fabric's color and print. It does not see measurements, the fabric's weight or drape, the inside construction, the back of the garment, or anything hidden by a pose, a bag or a crop. The first group becomes the choice of garment and options; the second group comes from you, through your measurements, your fabric and your corrections.
What does the AI read reliably from a photo?
Vision models are trained on enormous numbers of pictures of clothes, and they recognize the categories humans use for them. For a clear front view of a whole garment, these are read well:
| What it reads | How reliably | What it becomes |
|---|---|---|
| Garment type (skirt, dress, hoodie, trousers) | Very reliably | The garment template |
| Overall silhouette (A-line, straight, flared, boxy) | Reliably | Skirt or body shape options |
| Length relative to the body | Reliably when the hem is in frame | Length option |
| Neckline shape | Reliably from the front | Neckline option |
| Sleeve length and shape | Reliably when the arm is visible | Sleeve options |
| Visible closure (buttons, exposed zip, ties) | Reliably when in view | Closure option |
| Pockets | When visible; patch pockets more than in-seam | Pocket option |
| Gathers, pleats, ruffles, tiers | Usually, if the picture is sharp | Fullness and detail options |
| Fabric color and print | Reliably | The 3D preview's appearance |
| Woven or knit | Often, from how it hangs and the presence of ribbing | Fabric type default |
The output of this stage is a structured choice rather than a drawing: for a dress, something like "fit-and-flare, midi, square neckline, short puff sleeves, gathered skirt, zip at the back, no visible pockets". That choice is shown to you, and the pattern pieces are then drafted by a deterministic engine from blocks and your measurements, as how Sewly drafts a pattern explains. The AI's job ends at recognition.
What does the AI get wrong, and why?
Most misreadings come from information that is not in the picture rather than from the model misjudging what is.
- Fit. A relaxed top on a slim model and an oversized one on a larger model can look alike. The model reads the garment relative to the body it is on, and may call an oversized cut "relaxed". Check the fit option first.
- Length under a crop. If the photo stops at mid-thigh, the AI has to guess whether the skirt is mini or midi. It guesses something plausible and tells you; correct it.
- Hidden closures. An invisible zip is invisible to a vision model too. A dress with no visible opening is usually drafted with a zip at the back, which may or may not match the original.
- Gathers versus pleats, darts versus princess seams. At a distance, in a busy print or in soft light, these blur together. A close-up second photo helps.
- Pockets in a seam. In-seam pockets are, by design, hard to see. If the original has them and you want them, say so.
- Straps and necklines seen at an angle. A three-quarter view foreshortens the neckline and can turn a square neck into a scoop.
- Lining and facings. Nothing on the outside shows how the inside is finished. The pattern uses its own construction for the garment type.
Every one of these is a choice you can see and change before the pattern is drafted, which is why it pays to read the AI's summary rather than skipping to the preview.
Can the AI measure the garment, or you, from a photo?
No, and no tool should claim it. A photograph has no scale: a model's height, the camera's distance and the lens all change the apparent size of every part of the garment, and the fabric folds and stretches around a body you cannot measure through a screen. What a vision model can estimate is proportion, such as a hem that falls a little below the knee, and that is what it uses. The actual numbers, the centimeters that become the pattern, come from your tape measure.
That is also true of your own body. A measuring assistant can explain where each measurement is taken and check the numbers you type for plausibility; it cannot read them from a selfie. The guide to how to take your body measurements walks you through each one in ten minutes. Four numbers (height, bust, waist, hip) are enough to start, with the rest filled from the nearest standard size.
What does the AI know about the fabric?
It sees the surface: the color, the print and its scale, and whether the cloth looks matte or shiny, crisp or soft. That is enough for the 3D preview to show the garment in a plain color or with the print taken from your picture. It is not enough to know the fiber, the weight, the drape or the stretch, and those decide how the garment will actually behave. A photo of a slip dress in a fluid satin will give you a slip dress pattern; sewn in a stiff cotton, the same pattern makes a different garment.
So the fabric decision stays yours. Name the fabric you will use when you upload the picture, or set it afterward, because woven and knit fabrics are drafted with different ease and different seam allowances. When you want the look of the original without the exact cloth, choose a fabric of similar weight and drape rather than the same color.
What makes a photo easier for the AI to read?
You get a better first draft by feeding the AI a picture that shows what it needs.
- A front view of the whole garment, hem included, worn or on a hanger. A flat lay works too.
- A plain background and even light. A busy background and strong shadows add things the model has to ignore.
- The garment, not the pose. A hand in a pocket hides the pocket; a crossed arm hides the closure.
- A second picture of any detail you care about: the neckline up close, the back if you want the back copied.
- A screenshot rather than a link when the picture is on a page the tool cannot open, such as some Instagram posts. Screenshot to sewing pattern tips covers cropping and resolution.
- A sentence of corrections with the picture: "but knee-length, long sleeves, in linen".
The photo to sewing pattern page describes what happens from upload to PDF.
What happens after the AI has looked?
The AI hands over a set of choices, and the drafting engine draws the pieces from flat-pattern blocks with your measurements and the ease for the fit it chose. From that point on there is no AI in the loop: every option and every measurement can be changed by hand, and the pattern redraws instantly and identically each time. The 3D preview then shows the silhouette and proportions on a mannequin built from your measurements, in the original's print if you want it. It is a preview of shape, not a physics simulation of fabric, so a bias-cut hem will not drop and a jersey will not cling on screen the way they do in cloth. See it in 3D before cutting explains what the preview is for.
What will the AI not draft, whatever the photo shows?
Recognition and drafting are different limits. The AI will happily recognize a tailored blazer, a trench coat, fly-front jeans, a shirt with a collar stand, a bra or a swimsuit; at the time of writing (October 2026), the drafting engine has no blocks for them, and a fair tool says so and offers the closest of its 20 garments, such as a kimono jacket for a blazer or tapered pull-on trousers for jeans. Knitwear is another category entirely, since it is knitted to shape rather than cut from cloth. The full list, along with the other things that stay in the sewist's hands, is in what AI cannot do in sewing.