Fashion · UGC imagery
WorkflowFashionRecreating UGC-style fashion images with your own garments
The mirror selfie, the cafe-table flat lay, the street shot mid-stride. This playbook shows how to recreate the look of the shots your audience saves, with your own garments in them and a person approving every image.
Valnora & Elric team6 min read
- Format
- Playbook
- Steps
- 6
- Solution
- Workflow
- Parts used
- 4
Some of the best-performing fashion creative does not look like a campaign. It looks like a customer: a mirror selfie in a small bedroom, a kurta on a cafe chair, a mid-stride street shot with the phone slightly tilted. Shooting that look for every SKU takes time most D2C teams do not have.
This playbook shows how to recreate the shot, not the person, using your own garments in Agent Serve. You keep a reference image for pose and composition, add your product photos, generate a few variants with an image model, and a person approves what gets used.
What goes in
Three inputs make a shot.
- A reference image. It tells the model the pose, framing, angle and light. Use one you have the right to use, or one you shot yourself. It is a guide for composition only.
- Product photos. Clean packshots or flat lays of your garment, plus a close-up of any print, embroidery or logo. These are what the model should reproduce.
- A brief row. One row per shot in a Table: SKU, setting, an invented model description, aspect ratio and channel. The table becomes your shot list and your review log.
From a reference shot to an approved UGC-style image
Inputs · Brief
- 01Reference shotA shot you have the right to use, chosen for pose, framing and light only.
- 02Product photosPackshots or flat lays of your own garment, front and detail views.
- 03Shot briefOne row per shot: SKU, setting, model description, aspect ratio, channel.
Workflow · Generate
- 04Prompt writerTurns the brief row into a prompt using your house pattern.
- 05Image GeneratorRuns the pinned model with the reference and product photos attached.
- 06Library · Generated/ImagesEach image is saved with its prompt, model and cost.
People · Review
- 07Brand reviewChecks garment accuracy, print and logo, skin and hands, and the guardrails.
- 08Approved for useApproved images are marked in the brief table and labelled for AI use where required.
Pick and pin a model
In Chat, open the composer's + menu, choose Media models and pin an image model for that chat. It stays separate from the reasoning model that writes your prompts. In a workflow, the Image Generator block takes the same choice.
Models come from a live catalogue under More models, so treat these as examples:
- Seedream 5.0 for value, when you need many lifestyle variants and small detail drift is acceptable.
- Nano Banana 2.1 when print or logo fidelity matters, such as block prints, monograms or text on a tee.
- Others such as FLUX.2 or GPT Image 2 are worth testing on your own garments.
Run the same brief through two models before you commit. Garments vary: a plain linen shirt forgives a lot, a dense floral print does not.
How to measure it
US$0.03–0.15
Per image, illustrative range
Depends on model and resolution. The Library records the actual cost of each file.
4–6
Variants per shot, illustrative
Enough to pick from without paying for dozens of near-duplicates.
Illustrative example. Figures show how the workflow is measured; they are not customer results.
Write the prompt in a fixed pattern
A house pattern keeps results consistent across SKUs and people. One that works well:
- Shot. "Phone mirror selfie, camera at chest height, slight lean, natural window light from the left."
- Person. "An invented adult woman in her mid-20s, shoulder-length hair, relaxed smile." Never a named person or a description of the reference subject.
- Garment. "Wearing the kurta exactly as shown in the product photos: indigo block print, mandarin collar, three-quarter sleeves."
- Setting. "Small bedroom, wooden wardrobe, a plant in the corner, slightly messy bed."
- Finish. "Casual phone photo, slight grain, no text, no logos other than the garment's own."
Attach the reference image and the product photos as references (image-to-image). In a workflow, an Agent block can write this prompt from the brief row, and the Image Generator block runs it. Run the workflow per new product, or loop it over a list of brief rows. It does not batch a whole table in one step.
Review before anything is used
Every generated image is saved to the Library under Generated/Images with its prompt, model and cost, so a reviewer can see exactly how it was made. The reviewer checks the garment first: colour, print, cut, logo placement. Then hands, fabric folds and any text, which image models still get wrong.
What the solution produces
Shot brief row (illustrative)
Record- SKU
- KRT-0412 · indigo block-print kurta, size M
- Reference
- Mirror selfie, phone at chest height, natural window light, slight lean
- Model description
- Invented adult woman, mid-20s, shoulder-length hair, relaxed smile
- Setting
- Small bedroom, wooden wardrobe, plant in corner
- Model
- Nano Banana 2.1 (print fidelity needed)
- Format
- 4:5 for feed, 9:16 crop for stories
- Status
- Generated · 4 variants · waiting for review
Review checklist
Checklist- Garment colour, print and cut match the product photos
- Our logo is correct; no other brand marks or artwork appear
- The person is invented and does not resemble the reference subject
- Hands, fabric folds and text on garments look natural
- AI disclosure added where the ad platform requires it
Mark approved images in the brief table. Rejected ones stay in the Library with their provenance, which helps you tune the prompt pattern.
Guardrails
These are the rules that keep this safe for the brand.
- Copy the shot, not the person. Use references for composition only. Do not recreate a real person's face, body or likeness, including creators, celebrities and your own customers, unless you have their written permission.
- No other brands. Do not reproduce another brand's logo, signature print or distinctive artwork, even in the background.
- Label AI content. Meta requires advertisers to disclose digitally created or altered content in some ads, such as ads about social issues, elections or politics, and labels some AI content itself. Check the current policy for each platform you post to, and keep a disclosure field in the brief table.
- Keep spend visible. Admins set a cost confirmation threshold in Access policies (US$1 by default), so a generation estimated above it asks before running. A running generation shows a progress card with cancel.
Controls and governance
Roles and permission groups
Admin, write and read roles per workspace, plus permission groups that limit which models, blocks and integrations a team can use.
Run logs
Every workflow run records each block’s inputs, outputs and errors, so a reviewer can trace how a result was reached.
Audit trail
Proposals, approvals, declines and executed actions are written to the workspace audit log, which admins can export.
Bring your own keys
Use your organisation’s own model provider keys at organisation or workspace level.
Deployed versions
Workflows run from a deployed version, so a change is reviewed before it reaches production runs.
More on how Agent Serve handles access, data and audit on the security page.
Where it goes next
Approved stills make good first frames for short video ads; see product video ads. If they end up in paid social, running Meta ads from Agent Serve covers drafting and reviewing campaigns.
Questions teams ask
Can I upload a creator’s photo and put our dress on them?
No. Use a reference for pose, framing and light only, and describe an invented person in the prompt. Do not recreate a real person’s face or likeness, and do not copy another brand’s logo or distinctive artwork.
Which image model should I start with?
For everyday lifestyle shots, a value model such as Seedream 5.0 is a sensible start. When a print, embroidery or logo has to stay exact, try a model such as Nano Banana 2.1. The catalogue under More models changes, so test two models on the same brief before you settle.
Can it generate images for my whole catalogue at once?
Not as a single button. The workflow runs once per brief row, so you run it for each new product or loop it over a list of rows. Each run still saves to the Library and waits for review.