An on-model photo for a print-on-demand product is your flat mockup (the shirt, hoodie or tote with the print on it) handed to an image model as a reference, together with a person who is either fully generated or a model who signed a release, and a prompt that describes the pose, the place and the light. On 8frame the chain is upload, a reference image node, a written fix, an optional cut-out and an upscale. The limit people do not expect: the model redraws your print when it puts it on a body, so the design on the chest can come back as a slightly different drawing of your design, and it has to be checked against the original file before the photo goes on a listing.
This page covers stills for listings. For motion (fabric loops, worn clips, drop montages), our print-on-demand video guide is the next step.
TL;DR
| Step | Node or model on 8frame | Fixed price in credits | Input it needs |
|---|---|---|---|
| Mockup and print file in | Upload | none | PNG, JPG, WebP, GIF, AVIF or BMP, up to 50 MB |
| Generated person, if you need one | Nano Banana 2 or Seedream 5 Lite | 8 / 11 / 17 / 22 (0.5K to 4K), or 6 | a written description |
| On-model photo | Seedream 5 Lite / Nano Banana 2 / Nano Banana Pro | 6 / from 8 / 21 at 1K or 2K | mockup, print file and person as references |
| Print correction | Flux Edit (FLUX.2 pro) | 7 for a roughly 1 MP input, more for larger | the photo plus one instruction |
| Second edit model | Kontext Max | 13 | the photo plus one instruction |
| Cut-out for a plain listing image | Background removal | 12 | one image |
| Enlarge | BRIA Increase Resolution | 6, 2x or 4x | one image |
The board, node by node
Start from the CHANGE OF CLOTHES template on /workflows/change-of-clothes: it opens as a board you can duplicate and rewire. The steps below are the print-on-demand version of it.
1. Put three things on the board
- The flat mockup from your print provider: the product, its colour, the print in place.
- The print file itself, exported as PNG, on a transparent or white background. This is the reference the model needs for the design, and it is also what you check against at the end.
- The person. Either a generated person you make on the board in step 2, or a photo of a model who has agreed in writing that you may use their image for product photos, including AI edits. Do not use a photo of someone who has not agreed, and never a public figure.
2. Make the person once, then reuse them
If you generate the model, do it once and keep the result as a reference for the whole range. A shop that changes faces on every listing looks like five shops. Write the person as a short spec on a sticky note (age range, hair, build, style) and generate a few candidates; keep one. From then on, that one image goes into every on-model node as the person reference.
3. Generate the on-model photo
Wire three references into one image node: the person, the mockup and the print file. Seedream 5 Lite takes up to 14 references for 6 credits per image; Nano Banana 2, the default, takes up to 10 (8 to 22 credits by size); Nano Banana Pro is 21 at 1K or 2K for a hero image. Prompt the photograph, not the product: "the person wearing the heather grey t-shirt from the mockup, front view from the waist up, in a bright apartment with plants, soft daylight, the chest print exactly as in the print reference, centred and flat on the fabric."
Keep the framing waist-up or wider. A print is easier to check when the shirt is flat to the camera; side angles and crossed arms fold the design, and every fold is a place for the model to invent.
4. Check the print, then fix it
Open the result at full size (Original in the download menu) next to the print file. Check: every letter of any text, the number of elements, the colours, the size of the print against the shirt, and the position (a chest print that drifted to the belly is a different product). If one thing is off, fix it with one written instruction in Flux Edit (7 credits for a roughly 1 MP input, more for larger inputs) or Kontext Max (13): "make the print smaller and move it up to the chest". If the text is wrong, a new take is usually faster than an edit.
5. Make the listing set
For each design you typically need: one on-model hero, one flat or ghost image of the product alone (background removal, 12), and one detail of the print. The detail shot is the honest one: crop it from your print provider's mockup, not from the AI photo, so the buyer sees the real artwork.
6. Enlarge last
BRIA (6 credits, 2x or 4x up to 8192 x 8192) keeps the content; Clarity (13, always 2x on 8frame) is diffusion-based and can add detail, which is risky on fine print. Enlarge after every edit, never before.
Which model for which step
| Need | Model | What matters here |
|---|---|---|
| Person, mockup, print and a pose reference in one node | Seedream 5 Lite | up to 14 references, 6 credits per image |
| Quick takes across a range | Nano Banana 2 | the default; 8 at 0.5K, 11 at 1K |
| One hero image for the shop banner | Nano Banana Pro | 21 at 1K or 2K, 42 at 4K |
| Many references with a quality setting | GPT Image 2 | 1 to 10 references; high (default) 29 square, 20 other ratios |
| Fix position, size or a stray mark | Flux Edit, Kontext Max | 7 (1 MP input) and 13 |
We have not measured which model keeps a print most faithfully. On the same references, run two models and keep the take whose print matches the file.
Where it breaks
- Text in the design. Lettering is where changes are most likely: a missing letter, a swapped word, a new font. If your design is mostly text, generate the scene and composite the real print in your own editor, or show the real mockup as the main image.
- Colour of the garment. The model may lighten or darken the shirt. A buyer who orders "forest green" and receives a darker green has a reason to return it.
- Print scale. The print can come back larger or smaller than your provider will print it. Compare it with the provider's mockup, which shows the real print area.
- Folds and drape. Folds bend the design in ways a printed shirt would, but also in ways it would not. Flat front views are the safe default.
- Many colourways. Do not generate colourways you have not seen as a mockup from your provider. The model's version of "sand" is not your provider's sand.
Etsy's rules on AI and on who made it
If you sell on Etsy, two lines on Etsy's "What can I sell on Etsy?" help page apply to this workflow directly:
- AI disclosure. Etsy allows AI tools for certain items "as long as they contribute their own creative input and follow Etsy's policies", and "Sellers must disclose within their listing description if an item is created with the use of AI." Seller-prompted AI creations are part of the "Designed by a seller" category and "must disclose the use of AI."
- Production partners. "If you work with a production partner, you must disclose that production partner in your relevant listings." A print-on-demand provider that prints your design is the usual case.
Etsy's page is about the item. Whether an AI-made product photo needs its own note is not spelled out there, so the cautious reading is: say in the description that the photos are AI-generated on a model, and keep at least one image that shows the real product as the provider will print it.
If an image with a generated person is used in an advertisement, check the local rules on synthetic performers in every place the ad runs. This is general information, not legal advice.
The rights you need
You need the rights to the design (your own artwork, or a licence that covers products and AI processing), the rights to the face (a generated person, or a model's written release), and your print provider's permission to use their mockups in your own images if their terms say so. On 8frame, commercial use of what you generate is cleared on paid plans, and output carries no watermark on any plan.
FAQ
Can I make on-model photos without hiring a model? Yes, with a generated person you create once and reuse as a reference. If you use a real model's photo instead, get their written agreement to AI edits first.
Why does the print on the AI photo look different from my design? The model redraws the print when it puts it on a body. Check text, colours, scale and position against the print file at full size, and fix one thing at a time with Flux Edit or Kontext Max.
Do I have to tell Etsy buyers the photo is AI? Etsy's help page requires sellers to disclose AI use in the listing description when an item is created with AI, and to disclose production partners. For the photos themselves, saying they are AI-generated and keeping one real mockup image is the safe default.
Which 8frame model takes the most references? Seedream 5 Lite, up to 14 reference images, at 6 credits per image. Nano Banana 2 and GPT Image 2 take up to 10.
Should the main image be the AI photo or the mockup? That depends on the marketplace's main-image rules. Where a plain product image is required, use the provider's mockup or a cut-out (background removal, 12 credits) and put the on-model photo second.
Sources
- Etsy Help Center, What can I sell on Etsy? (AI creative input and disclosure in the listing description, seller-prompted AI creations, production partner disclosure), checked 2026-11-09
- 8frame prices, defaults 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.