Prompts and workflows for smart glasses and live visual AI

What it is good at

• Hands-free capture: you keep both hands on the fabric, the sample or the rail while you ask.
• The assistant sees the same thing you do, so you stop describing objects and start asking about them.
• Spoken notes taken in the moment beat photos you have to reconstruct three days later.
• Voice answers arrive while you are still standing in front of the thing you asked about, so you can correct it immediately.

Where it falls short

• Camera assistants guess at fibre content, weight and construction. Treat every material call as a hypothesis to confirm with a supplier, never a spec.
• Colour read through a camera is unreliable under mixed showroom lighting. Confirm against a physical reference or a measured value.
• Battery and heat limit continuous live-view sessions to short bursts, not a full market day.
• Recording people, showrooms and unreleased collections raises consent and confidentiality issues before it raises technical ones.
• Answers are conversational and easy to mishear. Anything that ends up in a purchase order needs re-reading in text.

How to use it for creative work

1. 01

   Decide what you are collecting before you put the glasses on

   Pick one job for the session: fabric shortlist, competitor pricing sweep, colour story, construction notes, or styling options. A session with one job produces notes you can sort afterwards. A session with five produces a voice memo nobody reopens.

2. 02

   Set the answer format out loud at the start

   Open the session by telling the assistant the shape you want every answer in, for example, three bullet points, under twenty words each, always ending with a confidence level. It carries that instruction across the following questions, so you stop repeating it.

3. 03

   Ask one question per object

   Point at the item, name it in two words, ask one thing. 'This knit, likely fibre blend and why.' Stacked questions produce a paragraph you cannot act on while walking.

4. 04

   Say the uncertainty rule once

   Tell it explicitly: when you are not sure, say 'not sure' and tell me what to check. Without this, a visual assistant will confidently name a fibre it cannot possibly see.

5. 05

   Capture the identifier, not just the impression

   For every item worth remembering, say the stand number, supplier name or shelf code out loud so it lands in the transcript. An opinion about a fabric with no way to find that fabric again is worthless.

6. 06

   Convert the transcript into a brief at a desk

   Back at a screen, paste the whole session transcript into a text assistant and ask it to produce a structured table: item, supplier, observation, confidence, next action. This is where a walkaround becomes a sourcing document.

7. 07

   Verify before anything becomes a decision

   Mark every line that came only from the camera. Those get confirmed with a supplier spec sheet, a hand feel or a measured colour value before they influence a buy, a costing or a client presentation.

Ready-to-use prompts

For the rest of this session, answer everything I ask in three short bullet points, under twenty words each, and finish with High, Medium or Low confidence. If you cannot actually tell from what you can see, say 'not sure' and tell me what I would need to check.

Look at this fabric. Give me your best guess at fibre blend, approximate weight and weave or knit structure, and say which of those you are least sure about and why.

Look at this garment. Describe the construction details a factory would need to quote it: seam types, closures, finishing and anything unusual. Flag anything that would add cost.

Describe the colours in front of me as a palette: name each one in plain language, say which is dominant, and tell me how the lighting here is probably distorting what you see.

Look at this rail. Summarise what this brand is betting on this season in three lines: silhouette, materials and price positioning. No adjectives you cannot justify from what you can see.

Turn what I am looking at into three moodboard keywords and one sentence I could paste into an image prompt later. Keep it concrete, surfaces, shapes, light, not mood words.

Look at this piece. Give me three ways to style it for three different customers, one line each, and say which one I should shoot first and why.

Read what you can see on this label or shelf ticket back to me as structured data: brand, product, price, any material or care information. If a field is unreadable, say unreadable rather than guessing.

This is a sample. Tell me the three things most likely to be wrong with it, then give me one recommendation: approve, approve with changes, or reject, and the single reason behind it.

Describe the light in this space the way a photographer would: direction, quality, colour temperature, and the one change that would most improve a product shot taken here right now.

Recap everything I asked you about in this session as a numbered list: item, where it was, what you told me, and how confident you were. Mark anything I need to verify before using it.

Questions people ask

Can smart glasses actually identify a fabric?

No, not reliably. A camera can see weave structure, surface sheen, drape and stitch density, and from those it can make a reasonable guess at a blend. It cannot see fibre content. Use the guess to narrow a conversation with a supplier, never to write a spec.

What is the difference between asking a phone and asking glasses?

Framing and hands. With glasses the assistant sees roughly what you see without you setting up a shot, and your hands stay on the product. The tradeoff is that you get an answer by voice, which is harder to check and easy to mishear.

Is it legal to record in a showroom or at a trade fair?

Often not, and it is usually a policy question before a legal one. Many fairs, showrooms and factories prohibit recording outright. Ask first, and assume unreleased collections and other people's samples are off limits unless you have written permission.

How do I stop the assistant inventing details?

Give it an explicit uncertainty rule at the start of the session and repeat it whenever it slips: if you cannot tell from what you can see, say 'not sure' and tell me what to check. Then ask for a confidence level on every answer so overconfidence is visible.

What should I do with the session afterwards?

Paste the transcript into a text assistant at a desk and turn it into a table of item, source, observation, confidence and next action. The walkaround collects raw material; the brief is written later, with verification.