A fashion brand asked me to help them produce their visuals with AI. They already use AI daily: they're not beginners. What they wanted was to move from isolated experiments to a method they control, with a tool that belongs to them.

I'm deliberately not giving their name or product. The case is typical of what I encounter with small brands, and that's what makes it useful.

What problem are we trying to solve?

A collection photo shoot costs time, organisation and logistics. Brands that turn to virtual models almost all hit the same difficulty: consistency. Getting a beautiful image is easy. Getting the same model, with the same face, the same body shape and the same style, across forty different images, is another matter.

Without a method, each image is a new roll of the dice. With a method, it becomes a piece of a whole.

Why not simply use a market tool?

Ready-made platforms exist, and they have their place. But they pose three problems for small brands:

you depend on their choices, their prices and their lifespan; what you've learned is tied to the tool, not to the skill; the settings, prompts and files stay with them.

My objective is the opposite: for the brand to leave with skills that live outside any tool. Whether they use Claude, GPT or Gemini tomorrow, they'll know what to ask them and how.

What's the method?

We build together a small application that runs on their computer. It needs no server or platform, and it doesn't depend on any particular AI assistant. It calls an image generation engine with an access key that the brand holds themselves.

The principle rests on four ideas.

Files rather than wishes. Each virtual model is described by a file: face, distinguishing features, body shape. The brand writes it, with AI help to find the right words. It's this file, not chance, that guarantees the model stays the same from one image to another.

A library of elements. Alongside the models, we store products, accessories, furniture, settings, jewellery and scenes. An image then becomes an assembly: this model, this garment, this setting. We never start from scratch.

Prompts that improve. The instructions sent to the image engine aren't written on the fly. They come from a base of proven prompts, which the brand adds to as they experiment.

Work done with them, not instead of them. I don't deliver a closed tool. I build with the brand, step by step, so they understand each piece and can modify it when their needs or AI models evolve.

What must we keep under control?

A few points require attention, and I advise every brand on them.

Transparency. Rules on disclosing AI-generated images are evolving. It's better to research and decide your own charter before publishing, rather than discover it afterwards. Brand consistency. A virtual model is part of visual identity. It's decided, documented and kept stable. Security, in the right measure. On a tool that runs on the computer of the person using it, we avoid excessive protections. The real point of vigilance is the access key: knowing where it's stored and who can read it.

What to take away from this

A small brand doesn't need another platform. It needs three things: clear files, a library of reusable elements and a way of talking to AI that doesn't depend on any particular tool.

If you recognise yourself in this situation, whether you're in fashion or another image-based business, this is exactly the type of work I do in workshops with SME leaders.