Ask a portrait painter what takes longest in the studio and many will not say the face. They will say the clothes. A collar decides where the eye goes; a sleeve decides the shape of the silhouette; a jacket that is the wrong black can flatten a whole canvas. Clothing has always been part of the image's grammar, from the lace ruffs of Dutch group portraits to the tailoring that told Sargent's viewers exactly who his sitters were. What is new is how cheaply that grammar can now be sketched.

The costume sketch, redrawn

Stylists, photographers and portrait artists have always made visual drafts before committing. A stylist pins swatches next to a test shot; a painter does a quick oil study; a photographer shoots Polaroids before the real roll. The newest version of that habit is to take one existing photograph and change clothes in a photo with AI, leaving everything else untouched.

That last part is what makes it useful. Earlier generative tools re-rendered the whole image, so the face would drift, the pose would soften and the background would quietly rearrange itself. For an artist that is useless: the point of a draft is to hold every variable still except the one being tested. The better tools now lock the person, the pose and the light, and redraw only the garments, either from a short list of presets such as a suit, a gown or a long wool coat, or from a written description when the idea is more specific, down to a particular colour, a uniform or a period costume.

Four places it earns its keep

Styling experiments. A photographer planning an editorial can run the same model through six outfit ideas in an afternoon, then take the two that work to the actual fitting. The drafts are the argument for which clothes to pull.

Period dress side by side. Costume designers and history-minded painters use it to set eras against each other on a single face: the same sitter in a 1920s drop-waist dress, a 1950s tailored suit and a 1970s wide-collar shirt. Seeing one person carry three decades is a quicker lesson in how silhouettes changed than a shelf of reference books.

Previewing a commission. Portrait artists often lose time to a question the client cannot answer in words: what should I wear for the sitting? A handful of drafts on the client's own photo turns that into a choice between pictures. Where the client has one specific garment in mind, a virtual try-on works the other way round, fitting a photo of that exact piece onto their image.

Teaching fashion history. In the classroom it becomes a comparison tool. Students dress a single reference photo in the clothing of different periods, then argue about what the change does to posture, class signals and mood.

Where it still falls short

It is honest to say what these drafts cannot do. Fabric physics is the weak point: silk, heavy wool and stiff brocade can come back looking suspiciously similar, and the way a skirt swings or a coat hangs from the shoulder is often approximated rather than understood. Extreme poses, crossed arms, figures cut off by the frame and busy patterns overlapping the body still trip the models up. And period accuracy is only as good as the description; a prompt for "Victorian dress" will happily mix decades unless someone who knows the difference checks it.

That is why the output should always pass under a human eye before it informs a real decision. Treat it as a pencil study with good lighting, not as a costume record.

A draft, not a replacement

None of this replaces the fitting room or the painter's judgement. What it removes is the cost of asking the question. A stylist, photographer or portraitist who once tested two ideas can now test ten, then spend the real time and money on the one that matters. Platforms such as CreateVision AI offer this as one tool among many photo editors, but for artists its value is narrow and old-fashioned: a faster costume sketch, made on the actual sitter.