AI Guides › Workbench
By Nigel Guy · 6 min read
Most people try an AI try-on by uploading one selfie and typing "put me in this dress". The result looks polished, so it feels like an answer, but the model has been left to guess your proportions, the garment's cut and what it should change. It often changes your face, your body and the dress all at once, and you end up judging a stranger.
The rule: give the tool two clean photos and one narrow instruction, change only the garment, then treat the output as a style preview, never as a fit check.
Check current plan limits and prices on each tool's own page. They change often, and none of the free routes below needed a £ price to try at time of writing.
| Tool | What it does | Cost | Best for | Catch |
|---|---|---|---|---|
| Google Try-On (Search, Shopping, Images) | Puts a listed top, bottom or dress on your uploaded photo | Free at time of writing | Retail listings you are already browsing | Only tops, bottoms and dresses; no lingerie, swimwear or accessories. Google says it does not guarantee actual fit |
| Gemini app image editing | Edits your photo using text and, in Nano Banana 2, multiple reference images | Free tier plus paid Google AI plans; check the £ price on Google's plan page | Garments from a shop that has no try-on button | Editing needs you to be 18 or over; download resolution is higher on a paid plan |
| ChatGPT Images | Edits an uploaded photo from a text description, with a Select tool to mark an area | Free tier plus paid plans; check the £ price at checkout | Fixing one area at a time after a first attempt | Image and upload allowances depend on your plan and can drop at busy times |
Two photos, always.
Add the one-line facts the picture cannot show: the size you would order, the length, the fabric.
Search for the dress on Google, open the product or image, and choose "Try it on". Upload a photo of yourself, or one you have permission to use. Google's help page says it needs you to be 18 or over, signed in, with Search Services History switched on, and that it was available in 17 countries including the UK at time of writing. Your original photo is saved for convenience. Delete it with the camera icon, then "Delete original photo". Generated try-ons can be deleted from the "Recently tried" tray.
Use it first. It costs nothing and it is built for this one job.
Use this in Gemini or ChatGPT when the shop has no try-on button. Attach both photos, then paste:
You are a careful virtual stylist editing a photo for a shopper.
Context: Image 1 is a full-length photo of me. Image 2 is a product photo of [GARMENT_NAME] from [RETAILER].
Garment facts: size I would order [SIZE]; length [LENGTH]; fabric [FABRIC].
Goal: show me wearing the garment from Image 2, so I can judge the colour, shape and style.
Steps:
1. Keep my face, hair, skin tone, body shape, pose, background and lighting exactly as in Image 1.
2. Replace only my current top/bottom/dress with the garment from Image 2, keeping its true colour, pattern, neckline and hem length.
3. Let the fabric hang the way [FABRIC] would, using the length I gave you.
4. Do not slim, lengthen, smooth or retouch my body.
Output: one edited image, the same framing as Image 1.
Before you answer, check that nothing other than the clothing has changed. If either photo is missing, cropped or unclear, or if a garment fact is missing, ask me for it instead of guessing.
Fill in the bracketed items. After the first result, correct one thing at a time ("shorten the hem to match the product photo") rather than rewriting the whole request.
Do this one dress per message, in the same chat, so the same you appears each time.
You are a careful virtual stylist. In this chat I will send up to [NUMBER_OF_GARMENTS] garments, one at a time, each with a product photo, after my full-length photo.
For each garment: edit my photo so I am wearing it, changing nothing except the clothing. Use the same framing, pose and lighting every time so the results are comparable.
Output for each one: the edited image, then a three-line note: what the colour and shape do for me visually; what you could not tell from the product photo (length, stretch, opacity); and what I should check in the retailer's size guide.
Garment 1 is [GARMENT_1_NAME], size [SIZE_1]. Wait for my next message before starting Garment 2.
Do not rank the garments or comment on my body. If anything is unclear, ask first.
Fill in the number of garments, names and sizes. Save each image with the product link beside it.
The trap is mistaking a flattering image for a fit result. The garment in the output is a redraw, so length, drape and stretch are the model's best guess. Google says outright that its images do not determine or guarantee fit. Check the retailer's size chart, reviews and returns policy before buying; the picture only helps with colour and shape. A second trap is silent retouching: compare the output to your original and notice whether your waist, legs or face have moved.
It will not tell you the size to order, how a fabric feels, whether it is see-through in daylight, or how it moves when you walk. It will not handle lingerie or swimwear in Google's tool. It will not give you a trustworthy before-and-after of your own body.