A footwear drop often has a timing problem that has nothing to do with the factory. The hero sample is in house, the photograph is approved, and the medial-side still already shows the last, the lace pattern, and the outsole logo the brand will defend in a meeting. Then the retailer asks for the black/gum, the bone/white, and a limited olive colorway — and those uppers are still on a boat, or they exist only as a lab dip.
If you generate a new shoe from a text prompt, you usually get a handsome runner that is not this runner. The toe box rounds off. The eyestay stitching becomes a generic overlay. A different outsole pattern appears because the model has seen that pattern next to "sneaker." That is the wrong job. The photograph already holds the last. What you need is a colorway restyle, not a new shoe.
This is a practical workflow for doing that in the browser, before you stand up a local inpainting stack or book another sample-photography morning.
Treat the last as the contract
Footwear image-to-image only works if you are explicit about what must not move. The last, the pattern pieces, and the hardware are the product. Upper color, midsole color, and outsole color are the campaign variables.
Write the split down before you generate:
- Keep: the last and toe shape, the medial/lateral panel layout, stitch count and path, lace count and eyelet positions, tongue shape, heel counter, outsole tread and logo, the camera height and angle, the crop, and any identifiably handmade or sample marks (a crease at the flex point, a slight pull at a stitch, a scuff you have already accepted in the hero).
- Change: the upper color and material read (mesh stays mesh, nubuck stays nubuck), the midsole color, the outsole and foxing color, and — if the brief really requires it — the lace color. Not a new silhouette "to make it look more current." Not a cleaner midsole. Not a different outsole because it looks more "performance."
If the brief quietly includes "make it more premium" or "give it campaign energy," the model will often "improve" the shoe. That improvement is how a real sample becomes a cousin of itself: sharper logos, a more aggressive tread, a toe that no last in the line actually has.
One change family per run is enough. Do not ask for a new upper color, a new midsole, and a new studio sweep in the same prompt. Colorway first, on the original light. Environment later, from a keeper that still matches the last.
Start with a reference that can survive a restyle
Not every beauty shot is a good source. A shoe shot from too high, with the medial side in deep shade, or with the outsole logo half cropped, will not give the model enough structure to copy. You want a still in which the silhouette is unambiguous and the materials are readable.
A usable source usually has:
- A clear view of the last from a consistent catalog angle — often a 3/4 medial or a straight lateral, not a stylized "floating in mid-air" crop that hides the heel.
- Readable materials. If the mesh already looks like smooth leather in the source, the restyle cannot invent the mesh back.
- A quiet background. A seamless or a simple floor is easier than a lifestyle scene with gravel and plants the model will want to redesign.
- Enough resolution that the outsole logo and the stitching are actual shapes, not a smudge. A soft phone snap of a sample on a desk will drift faster than a locked catalog still.
If you only have a lifestyle shot, consider whether you are really asking for a colorway or for a new photograph. Image-to-image can recolor an upper in a street scene, but it will also try to "help" the pavement, the socks, and the cuff of the trousers. For a first colorway set, a clean pack still is the more honest source.
Write the prompt as a change list, not a mood board
"Make a premium olive sneaker, cinematic, product photography, 8k" is a mood board. It invites a new shoe. A change list names what stays and what moves.
A prompt that has worked for this job looks more like:
Same shoe and last as the reference. Keep the exact toe shape, panel layout, stitching, lace count, tongue, heel counter, and outsole tread and logo. Change only the upper to muted olive nubuck in the same panels, with a gum midsole and a dark gum outsole. Do not redesign the silhouette. Do not add or remove logos. Keep the same camera angle, crop, and lighting.
Name the material. "Olive sneaker" is how you get a generic mesh in a color the brand does not use. "Muted olive nubuck, same panel layout" is closer to a lab dip.
Name the parts you are willing to recolor, and stop there. If the midsole should stay raw white, say so. Models like to "complete" a color story by tinting the midsole to match the upper. That is a different SKU.
Avoid words that sound like a redesign: new model, updated, chunky, sleeker, more athletic, cleaner lines. Those are invitations to change the last.
A browser pass is enough for a first colorway set
You do not need a local node graph to try three unreleased colorways of a shoe you have already photographed. A browser-based AI image to image tool is enough for this kind of restyle: upload the catalog still, write a change list that protects the last, generate a short set of variants, and download the ones that still look like the same sample. Multiple models are available in the same workspace; if one pass invents a new outsole, try a different model on the original still rather than stacking a "fix" on a shoe that has already drifted.
Stay conservative with file prep. A sharp JPG, PNG, or WebP of the approved hero is better than a heavily sharpened export. If the source is within a normal upload size, do not pre-upscale it in a way that invents stitch texture the camera never saw. Generate a small batch — a handful of variants is enough to see whether the last is holding — and pick from those before you spend another pass.
The point of the browser pass is not to replace the later pack shot of the real olive sample. It is to give merchandising, the retailer, and the campaign team a colorway they can review while the sample is still in transit.
Check the last, then the logo, then the light
Review in a fixed order, or you will approve a pretty olive shoe that is slightly the wrong shoe.
- Last and pattern first. Overlay the output on the source if you can, or flip between them at the same zoom. The toe spring, the throat opening, and the heel line should not have moved. If the shoe has become longer, shorter, or more "fashion-forward" in the toe, discard the frame. Do not argue yourself into keeping it because the color is right.
- Stitching and hardware. Count the eyelets. Follow one stitch line from the toe to the midfoot. A new decorative stitch, a missing bartack, or a lace that now has seven holes instead of six is a failed restyle, even if a shopper might not notice at thumbnail size. Catalog teams notice.
- Logos and outsole. The outsole tread and any side logo should still be the ones on the sample. "Similar" is not enough for a line sheet. If the logo has become a vague chevron, start again from the original still.
- Material read. Nubuck should not come back as patent. Mesh should not come back as smooth leather. If the color is right and the material is wrong, the SKU is still wrong.
- Light last. Do the highlights on the midsole and the instep still agree with the original key light? A second sun on the toe, or a midsole that is now lit from the opposite side, is a common giveaway in a side-by-side grid.
If only the lace color is wrong, you can usually run a second, narrower pass from the best variant. If the last has already drifted, go back to the original still. Do not stack restyles on a shoe that is already a little bit wrong. Each generation will "help" it further.
A simple four-look set from one still
A practical set for a line sheet or a retailer preview, all from the same approved hero:
- Original color, proof pass. Run the source through the same prompt structure with "keep the original colorway" as a control. If the model cannot reproduce the shoe it was given, do not trust it with olive yet.
- Black / gum. Dark upper, gum midsole and outsole, same panel layout. Watch that the model does not also black out the midsole, and that the gum does not dirty the upper.
- Bone / white. A light upper is harder than it looks: the model will try to overexpose the midsole and lose the edge of the last against a light sweep. Keep the original lighting and the original background value.
- Limited olive. Muted olive upper in the specified material, with the midsole and outsole colors the brief actually named — not a full monochrome olive shoe unless that is the SKU.
Keep the crop and the camera height constant across the set. A buyer comparing four thumbnails will notice a zoom-in or a slightly rotated shoe faster than they will notice a slightly different lace tip.
When to leave the browser
Graduate to a local stack when the outsole logo or a woven label must stay pixel-accurate, when you are matching a locked e-commerce template, or when you are delivering dozens of colorways and must reuse a mask. Until then, a careful source still, a written keep/change list, and a short browser series will usually get you a usable colorway set without another studio call.
The test is simple: could the product-development person still recognize their sample? If yes, the restyle is doing its job. If you have to explain that it is "the same last, we just cleaned the toe a little," the shoe has already moved, and you should start again from the original still.

