Independent fashion labels live on tight calendars.
A collection moves from inspiration to samples to lookbook to launch in a matter of months, often with a team of three people doing the work of twelve. Every stage needs visuals, and most of those visuals are needed before the clothes physically exist.
That is the gap where AI image generation earns a place in a small studio. Not as a replacement for photographers, models or stylists, but as a sketchbook that can talk back. Google’s Nano Banana 2, one of the Gemini image models, has become a useful option here because it handles two things fashion teams care about: realistic fabric and lighting, and printed text that is spelled correctly.
Below is how it can fit into each stage of a collection cycle, along with the places where it should stay out of the way.
Stage one: mood and direction
Most collections start with a wall of torn magazine pages, fabric scraps and photos from a trip. The problem is that a mood board made of other people’s images can only point at an idea. It cannot show the exact combination you have in your head.
Image generation helps close that gap. A designer can describe a specific mood, then iterate quickly:
Fashion editorial portrait, medium close up on an 85mm lens. A woman in a sculptural tangerine raincoat with a high funnel collar stands against a hot pink paper backdrop, one hard beauty dish from camera left, crisp specular highlights, real skin texture, magazine cover color.
Change the coat to a boiled wool cape, swap pink for slate grey, and you have three directions to argue about in the morning meeting. Asking for up to four images in one run makes those comparisons fast.
Stage two: palette and material tests
Once a direction is chosen, the questions get practical. Does that burnt orange read as rust or as pumpkin under studio light? Does the quilted panel look expensive or bulky in a three quarter shot?
This is where reference images become important. Nano Banana 2 accepts up to 14 reference images, so a team can photograph actual swatches, a toile or a finished sample and ask the model to place them in a new scene. The output is not a substitute for a fitting, but it is a cheap way to rule out bad ideas before ordering a full roll of fabric.
Stage three: trims, tags and packaging
This is the stage where the newer Gemini image models genuinely change the workflow. Hang tags, woven labels, garment bags, mailer boxes and tissue paper all carry words, and older image generators were notorious for producing gibberish lettering.
The fix is simple: put every word that must print in double quotes, and say where it goes. For example:
Flat lay product photograph on cream linen. A heavyweight card hang tag with a cotton string loop, the brand name “ALDER & WREN” in a thin serif across the top, “AUTUMN / WINTER” beneath it, and “100% MERINO WOOL” in small capitals at the bottom. Soft window light, visible paper texture, crisp legible typography.
In testing across product labels, the Nano Banana tiers printed quoted brand and product names letter for letter. That makes it possible to mock up a full packaging suite for a buyer presentation without waiting on a print proof. Final artwork should still be set properly by a designer, with real brand fonts, before anything goes to the printer.
Stage four: lookbook planning and line sheets
Lookbook days are expensive. Location, photographer, hair and makeup, models and assistants all cost money per hour. Arriving with a clear shot list saves a lot of it.
Many studios now draft their shot list visually. A generated frame for each look, matching the planned lighting and framing, gives the photographer and stylist a shared reference. It also helps decide aspect ratios early: 4:5 for Instagram feed posts, 9:16 for Stories and Reels, 3:2 for a web banner. At 4K, Nano Banana 2 renders 3712 by 4608 pixels at 4:5, which is large enough for a printed line sheet mockup.
I run these drafts through PixelDojo, where Nano Banana 2 sits in one prompt box next to Nano Banana Pro and the lighter Nano Banana 2 Lite. The practical routine is to draft cheaply at 1K (2.25 credits per image on Nano Banana 2, or 2 on Lite), then render the chosen frame at 4K for 4.5 credits.
Where AI should stay out of fashion work
The industry has good reasons to be cautious, and a responsible studio should set its own rules. A few that make sense:
- Do not sell clothes with generated photos of clothes. Customers need to see the real garment, on a real body, with real drape and colour. Generated images belong in planning, not on product pages.
- Do not replace the people who give a brand its face. Models, photographers and stylists bring judgment and identity that a prompt cannot. Use AI to make their day more focused, not to cancel it.
- Do not imitate other designers. Prompting for a famous house’s signature look is both a legal risk and a creative dead end.
- Disclose. If a generated image is used publicly, for a teaser or a mood post, say so.
- Represent people deliberately. Describe age, skin tone, body type and hair specifically, and look critically at what comes back.
A sketchbook, not a studio
Fashion has always borrowed new tools early, from digital printing to 3D sampling. Image models like Nano Banana 2 are best treated the same way: a faster sketchbook for the uncertain early weeks of a collection, and a mockup table for the details buyers want to see. The clothes, and the people who make and wear them, are still the point.
