How to Create AI Product Photos with a Model

Choose the route before preparing images

A convincing generated image does not establish actual garment fit, real-world product dimensions, or evidence that someone used the product. Decide which of those facts your final image needs to communicate before you generate it.

Step 1: Set a simple, visible interaction

For a first attempt, make the product easy to inspect. A front-facing bag held at waist height is easier to evaluate than the same bag swinging behind a walking model. A bottle held beside the body exposes more of its label than one covered by two hands.

Write a short shot brief with four elements:

• The product and the details that identify it.
• The person’s relevant appearance or the role of the person reference.
• A pose that leaves the product visible.
• The intended crop and background.

For the tote example below, the goal was a three-quarter studio portrait of a fictional adult woman holding the bag by both handles. The pocket needed to face the camera. A warm-gray background and soft catalog lighting kept attention on the interaction.

Do not start with a complicated walking pose, changing outfit, new hairstyle, crowded setting, and tiny visible product all at once. Those changes make it difficult to identify which part of the result needs repair.

Step 2: Prepare separate product and person references

Use a sharp product image that shows its identifying details. Remove packaging clutter and avoid deep shadows over logos, seams, closures or important edges. If you have multiple views, keep their colors and construction consistent. Two conflicting versions of a product create an ambiguous brief.

The product source in our existing example is a fictional cream canvas tote. Its indigo handles and front pocket, brass snap, stitching and rectangular body provide visible checkpoints.

Choose a person reference with the face and hairstyle clearly visible. If the target is a full-body pose, a headshot alone leaves the body, clothing and stance for the generator to invent. A reference that already resembles the desired pose reduces that ambiguity.

Here, the person source was also synthetic: a fictional adult woman with short curly dark hair and a blue collared shirt.

These sources make the task reproducible without implying that a real model endorsed or handled a physical product. For a real campaign, use person and product materials you are entitled to use and confirm the commercial permissions of the selected model, feature and plan.

Step 3: Upload the references in Dreamina

Open the Dreamina AI image generator and choose an available image model and image-reference workflow. Clear any references left from another job before adding the new ones.

Upload the person and product as separate references. Check the thumbnails rather than assuming the file picker selected the right images. Keep the roles explicit: one reference supplies the person; the other supplies the product. A style photograph should not silently become a replacement product reference.

Choose the aspect ratio for the intended composition. A portrait crop works well for the waist-height tote interaction. Leave enough room for the hands and the entire bag; a tightly cropped face reference does not mean the output must be a headshot.

The existing example used two references, Seedream 5.0 Pro, a 4:5 ratio and the 2K setting. It was generated on October 20, 2026, under the account conditions recorded for that run. Those settings describe this example, rather than the default model or entitlement of every account. Check the model selector and generation cost shown in your own workspace.

Step 4: Describe the interaction and the details to retain

An effective prompt explains relationships between the inputs. “Fashion model with a bag” gives little guidance about which model, which bag, where the hands belong or which details matter.

This was the prompt used for the example:

Show the same fictional adult woman in her blue shirt naturally holding this exact canvas tote by its two indigo handles at waist height, with the front pocket facing the camera. Three-quarter studio portrait, neutral warm-gray backdrop, soft fashion catalog light, relaxed believable posture. Preserve her face and hair, the bag shape, cream canvas, indigo pocket, brass snap and stitching. Correct hand-handle contact, no extra fingers, no text.

“Exact” in a prompt expresses a requirement; it does not certify that the output satisfies it. The reference comparison in the next step is still necessary.

For another item, replace the interaction and the identifying details rather than copying the tote prompt unchanged. A cosmetics bottle might need the label fully visible and the fingers below the brand name. Earrings need a useful view of the fastening and shape. A backpack may need a side or rear pose, with strap attachments visible.

Generate a manageable set of candidates, then inspect them before spending more credits. If the interface offers variations, use them to compare a small number of poses rather than accumulating outputs with no acceptance rule.

Step 5: Check person, product and contact separately

The resulting tote image retained recognizable facial and clothing cues from the fictional person, and the bag’s cream body, indigo pocket and handles remained recognizable. The person held both handles, and the pocket faced the camera.

The pose, folds and view of the bag were redrawn. This is an example of a usable-looking generated interaction, not proof of unchanged product pixels, exact size, physical load capacity or reliable virtual try-on. The blue shirt came from the person reference; this run did not test putting a new garment on her.

Use three passes when reviewing your own output:

Inspect the native image at a useful zoom. A phone-sized preview can conceal a misspelled label or a broken handle connection. Check the product itself before deciding that attractive lighting makes the image ready for a listing.

Step 6: Repair the specific problem or change the route

If the face drifts but the product is acceptable, strengthen the person reference and simplify changes to hair, clothing and pose. If the bag is too small, describe its placement relative to the waist and hands; do not infer accurate centimeters from the generated result.

If fingers cover the label, change the pose so the hand supports the base or holds a noncritical edge. If hands repeatedly become implausible, choose a tabletop presentation or a model standing beside the product. That changes the interaction while keeping the product useful to the reader.

If the product changes, identify the changed detail by name. “Keep everything the same” is less actionable than “retain the single rectangular indigo front pocket and its centered brass snap.” A supported local editing workflow may help with a small area, but inspect the rest of the image after each edit. A request to change one region is not a guarantee that other pixels remain identical.

For products whose label, geometry or color must remain exact, preserve the original product layer in a photo editor and build the surrounding scene around it. Match perspective, light direction and contact shadow. If the desired view exposes surfaces absent from the source, obtain another product photograph or use a physical shoot; an AI reconstruction cannot supply verified hidden details.

Before approving an ecommerce asset, use the same comparison process described in checking AI product-photo fidelity.

For clothing: choose new-model generation or virtual try-on

The held-tote workflow is useful for bags and accessories. Apparel adds a different problem: the garment must be rendered around a body, with plausible sleeves, openings, folds, closures and occlusion. Use the route that matches your starting point.

FASHN Product-to-Model: create a new wearing image

FASHN’s documented Product-to-Model workflow accepts product inputs such as flat lays, ghost-mannequin images and hanger photographs. It also covers product-only accessories. In the Agent, add the product images, describe the model and shot, and optionally add face, pose or background references. The workflow supports up to six input images and up to four output variations.

Review the variations for garment construction before continuing to upscale, edit or reframe. A model reference guides a newly generated image; it is a different task from retaining every aspect of an existing model photograph.

This is an official documented alternative, not a garment test performed for this article. Its advertised preservation goals do not replace checking your own SKU details.

FASHN Try-On: keep an existing person and change the garment

For a selected person photograph, open Try-On, add the target model image, and upload the garment as the product input. Add optional styling instructions, choose the available output settings and variation count, then run the generation. Review the results in the Gallery before downloading or opening further editing tools.

The Try-On guidance favors a visible outfit area, a natural three-quarter or full-body pose, and clothing that does not obscure the body excessively. Hair covering the garment and extreme poses make the transfer harder to judge. Its product guidance favors on-model images, then suitable ghost-mannequin or flat-lay sources; it advises against hanger or floor images for this particular route.

That distinction matters: an input accepted by Product-to-Model is not automatically the best source for Try-On. A catalog garment image also cannot demonstrate how the physical item will fit a particular customer.

Photoroom AI Fashion Models: build a clothing image through staged controls

In the Photoroom web app, choose the correct Space, open Explore, and select AI Fashion Models; use See all if the tool is not initially visible. Add the clothing image and any additional outfit images needed for the intended look.

Set the available model, pose, background, size and quality options. Add a prompt or brand style where useful, then generate using the AI credits required by the workspace. Review the result, open it in the editor if needed, and download it or keep it in Designs in that Space.

These steps come from the current official web guide, rather than a new hands-on comparison. Review plan access and AI-credit costs before generating. Do not assume that a free account includes every model, output size or commercial permission.

Step 7: Export the approved image for its channel

Download the native result rather than taking a screenshot of the preview. Open the downloaded file and check the person, product and contact points again. Confirm the actual dimensions and crop; the chosen ratio alone does not establish the exported pixel size.

Prepare channel crops from the approved master. A square marketplace image and a tall social creative may need different framing. Keep the full product, hands and important label information inside the crop. Add exact promotional text in a design editor after approving the image, where it remains editable.

Retain the product source, person reference, prompt, chosen result and the record of changes. For an actual product listing, a reviewer who knows the SKU should approve details that a generic visual check cannot establish. Follow the destination’s current rules for AI-generated or model-led imagery.

Frequently asked questions

Can I create model product photos without hiring a model?

You can generate a synthetic person or use an authorized person reference to create a model-led image. That can support a concept or a suitable commercial image when the selected model, feature and plan permit the use. It does not establish a real person’s endorsement or prove physical product fit.

Can I use a single product photograph?

Yes, for a simple view, but unseen sides remain uncertain. Additional consistent views can clarify construction. If the final image depends on verified details behind the product or under a fold, supply those views or use a physical shoot.

Is a reference image enough to keep every product detail unchanged?

No. A reference guides generation, while the resulting product may be redrawn. Compare the original and output. For exact source pixels, retain the original product layer and use compositing or suitable background editing.

Can Dreamina create clothing-on-model images?

Dreamina can create images guided by garment and person references, subject to the selected model and available controls. The actual example here demonstrates a held tote. For transferring a garment onto an existing person, evaluate a dedicated try-on route and inspect the specific garment result.

Start with Dreamina when you need a clear person-and-product composition from separate references. Keep the interaction simple, approve the product details before the aesthetics, and move to a garment-specific or layered workflow when the task requires those controls.