To turn a flat lay into an on-model photo with AI, put the flat lay (front and back) and a photo of the model on one board and run an image model that takes several references, asking it to dress the person in the garment; for a ghost mannequin image, ask an edit model to shape the same garment as if worn by an invisible body. On 8frame that is Nano Banana 2 (8 to 22 credits per image by resolution, up to 10 references) for the on-model shot and Flux Edit (7 credits for about a 1 MP input) or Kontext Max (13) for the edit. The limit people do not expect: the model has never seen the garment on a body, so drape, length and fit in the result are its guess, and the ghost mannequin's inside neck and back are invented outright. Treat both as presentation images, and keep a real photo wherever fit is the reason to buy.
This page builds the product page set (on-model, ghost mannequin, detail) from one flat lay. For try-on tools aimed at shoppers see AI virtual try-on; for clothing video see clothing product video for an online store.
TL;DR
| Step | Node or model on 8frame | Price in credits | Input |
|---|---|---|---|
| Clean flat lay | Background removal | 12 per image | Your flat lay photo |
| On-model shot | Nano Banana 2 | 8 / 11 / 17 / 22 per image (0.5K / 1K / 2K / 4K) | Garment front and back, model photo |
| On-model, more references | Seedream 5 Lite | 6 per image | Up to 14 references |
| Ghost mannequin | Flux Edit (FLUX.2 pro) | 7 for about 1 MP, rises with input size | The flat lay or a mannequin photo |
| Ghost mannequin, alternative | Kontext Max | 13 per image | Same |
| Detail crops for zoom | BRIA Increase Resolution | 6 (2x or 4x) | The approved image |
| Larger zoom files | Topaz Image Upscale | 11 per started 24 MP block | The approved image |
The board, node by node
The closest live template is CHANGE OF CLOTHES on 8frame workflows, which changes the outfit on a person in an image. Open it as a starting board, swap its inputs for your flat lay and model photo, then add the ghost mannequin and detail steps below.
1. Shoot the flat lay for the AI, not for the page
The flat lay is now a reference, so shoot what the model needs to see: the whole garment, front and back, flat and unwrinkled, on a plain surface, in even daylight-balanced light, square to the camera. Add one close-up of the fabric and one of any print, label or hardware. Colour matters most: put a grey card or a white sheet in one test frame so you can check colour later. Run Background removal (12 credits) on the front and back to give the model clean garment shapes.
2. The model photo
Use a photo of a real person who agreed in writing to appear in AI-generated product images, or generate a person who does not exist. Full-length or knee-up, neutral pose, arms slightly away from the body, plain fitted clothes, the same light direction as the flat lay. A pose that hides the waist hides the part of the garment shoppers check.
3. The on-model shot
Attach the garment front, the garment back, the fabric close-up and the model photo to a Nano Banana 2 node. Prompt plainly: "the person wearing this exact garment, front view, studio white background, natural fit, same colour and print as the reference". Nano Banana 2 opens at its default resolution; 1K is 11 credits, 2K is 17, and it returns up to 4 images per run, so ask for four and pick. Seedream 5 Lite (6 credits) takes up to 14 references if you want to attach more angles. GPT Image 2 takes 1 to 10, at 20 credits for a non-square image at its default high quality.
Then check the take against the flat lay, line by line: neckline, sleeve length, hem length, number of buttons, pocket position, print scale, colour. Discard takes that changed any of them. A model that adds a pocket has made a different product.
4. The ghost mannequin image
A ghost mannequin (or invisible mannequin) shot shows the garment in a worn shape with no body. There are two AI routes:
- From a real mannequin photo. Photograph the garment on a mannequin, then ask Flux Edit or Kontext Max to remove the mannequin and fill the inside neck. Only the hidden inside is invented; the outside is your photo. This is the more honest route.
- From the flat lay. Ask the edit model to shape the flat garment as if worn by an invisible body. Everything about the 3D shape is invented: shoulder slope, drape, how the hem falls. Use it for simple shapes (T-shirts, sweatshirts) and check seams and labels closely.
Flux Edit costs 7 credits for about a 1 MP input and rises with the input size (up to 87 for very large files), so downsize a 40 MP camera file before you edit. Kontext Max is a flat 13 and can match the input image's ratio.
5. Detail shots and zoom files
For the zoom images, crop the approved on-model or ghost image, or use your real fabric close-up. To enlarge, use BRIA Increase Resolution (6 credits, 2x or 4x, content-preserving, up to 8192 x 8192) or Topaz (11 per started 24 MP block). Avoid Clarity (13) for fabric: on 8frame it always enlarges 2x and it is diffusion-based, so it can add weave and texture detail that is not in the garment.
Which model for which step
| Model | References | Ratios with 4:5 | Price | Fits |
|---|---|---|---|---|
| Nano Banana 2 | up to 10 | yes | 8 / 11 / 17 / 22 | On-model shots, 4:5 product images |
| Nano Banana Pro | up to 10 | yes | 21 at 1K and 2K, 42 at 4K | On-model shots where you want the Pro model |
| Seedream 5 Lite | up to 14 | no | 6 | Many references, square or portrait 3:4 |
| GPT Image 2 | 1 to 10 | no | 29 square, 20 other ratios (high) | When the label text must be legible |
| Flux Edit | edits one image | matches input | 7 for about 1 MP | Mannequin removal, background clean-up |
| Kontext Max | edits one image | 13 ratios incl. 4:5, or match input | 13 | Ghost mannequin, ratio change |
| BRIA | upscale | keeps input | 6 | Zoom files that must not gain detail |
We have not measured which model keeps a print or a colour closest to the garment. Choose by references and ratio, and compare every take with the flat lay.
Where it breaks
- Fit is a guess. The flat lay does not tell the model how the garment hangs on a body. Length, drape and tightness in the result can be wrong even when the garment looks right. If fit sells the item (tailoring, denim, swimwear), photograph it on a person.
- Small details change. Buttons, zips, labels, embroidery and print scale are where takes drift. Check each; regenerate rather than publish.
- Invented insides. The inside back and neck of a ghost mannequin image from a flat lay is invented. Do not show a size label you did not photograph.
- Colour. Generated colour can shift. Compare against the real garment under daylight and against your colour test frame. Make each colourway from its own flat lay, not by recolouring one.
- Upscaling texture. Clarity is diffusion-based and can add detail that is not in the source. Use BRIA or Topaz on fabric.
- No vector output. Size charts and logos for the page come from your design files.
Where the images go: marketplace rules
Google Merchant Center. Google's image_link rules (checked 2026-11-09) ask for at least 500 x 500 pixels from 31 January 2027, recommend around 1500 x 1500 or above, cap files at 64 megapixels and 16 MB, ban promotional overlays, watermarks, borders and added logos, and say "Submit a unique image for each variant" and "Show only one variant per image". They also say "All images created using generative AI must contain meta data indicating that the image was AI-generated": keep the metadata your tools write.
Amazon. Amazon has its own image rules for clothing listings. Read the current clothing style guide for your marketplace in Seller Central before you upload a ghost mannequin or on-model main image.
Your own store. You set the rules, but the buyer still compares the photo with the parcel. An on-model image that flatters the fit is a return.
Competitor tools package the same steps: Flora's techniques page lists a "Ghost Mannequin System", "Virtual Try-On" and a "PDP Set". The difference on a board is that every step is a node you can inspect, rerun and price.
FAQ
Can AI turn a flat lay into an on-model photo?
Yes: attach the flat lay front and back and a model photo to an image model with several references, such as Nano Banana 2 (up to 10). Check neckline, lengths, buttons, print and colour against the garment.
How do I make a ghost mannequin photo with AI?
Photograph the garment on a real mannequin and remove the mannequin with an edit model (Flux Edit 7, Kontext Max 13), or shape a flat lay into a worn form. The first route invents less.
Is an AI on-model photo accurate for fit?
Not reliably. The model guesses how the garment hangs. Use real photos where fit is the reason to buy.
Which upscaler should I use for fabric detail?
BRIA Increase Resolution (6 credits, 2x or 4x) or Topaz (11 per 24 MP block). Clarity is diffusion-based and always 2x on 8frame, and it can add texture that is not there.
Can I use one AI image for every colour?
Not on Google Merchant Center, which asks for a unique image per variant showing that variant. Make each colourway from its own flat lay.
Sources
- Google Merchant Center Help, Image link [image_link] (500 x 500 minimum from 2027-01-31, around 1500 x 1500 recommended, 64 MP and 16 MB limits, unique image per variant, AI metadata, no overlays or borders), checked 2026-11-09
- Flora, Techniques (Ghost Mannequin System, Virtual Try-On, PDP Set), checked 2026-11-09
- 8frame prices, reference limits, ratio settings, upscaler behaviour and limits: the 8frame app, checked 2026-11-09
Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.