Skip to content

Self-serve signup opens soon — book a demo for early access.

Book a demo
Book a demo
← All guides Guide

Getting Started with Brand-Trained Models

How to upload reference images, train a private visual model on your brand aesthetic, and generate your first on-brand product photography.

A brand-trained model is the foundation of consistent AI product photography. Unlike general-purpose image generators that produce a statistical average of the internet, a model fine-tuned on your reference images learns your specific fabric behaviour, casting language, lighting signature, and editorial mood.

This guide walks through the end-to-end workflow: preparing reference assets, training your model, and generating your first production-ready images with Photo Wizard.

Why brand training matters

Fashion brands invest heavily in visual identity. That identity lives in details — how jersey drapes versus structured tailoring, how skin is lit, how models relate to camera. Generic AI tools approximate “fashion photography” but not your fashion photography.

Brand training encodes those patterns into a private LoRA (Low-Rank Adaptation) layer on a production base model. Every module on the platform — Photo Wizard, Flat Lay, Cut-Out Photo, Hero to Video — can route through that model, so output stays on-brand across channels.

Step 1: Prepare your reference set

Quality of training data determines quality of output. Aim for 30–80 curated images that represent your brand consistently.

What to include:

  • On-model campaign and lookbook photography from recent seasons
  • Packshots and flat lays that show garment construction clearly
  • A mix of full-body and detail crops
  • Consistent casting and lighting where possible

What to avoid:

  • Heavy watermarking or third-party retail photography you do not own
  • Extreme filters that distort colour science
  • Single-outlier images that do not match your usual aesthetic
  • Logos or licensed third-party prints without clearance

Organise files by category if your brand spans multiple (womenswear, menswear, accessories). The platform reads your workspace brand_config to adapt prompts — categories and gender focus should be set accurately in workspace settings.

Step 2: Upload and configure Brand DNA

In your workspace, navigate to Custom Models and start a new training job.

  1. Upload reference images — drag-and-drop or bulk upload. Supported formats include JPEG, PNG, and WebP.
  2. Name your model — use a version label (e.g., “SS26 Core”) so you can retrain each season without losing history.
  3. Review Brand DNA summary — the platform analyses your uploads for dominant patterns: palette, lighting ratio, casting register, and compositional habits. Adjust the summary if it mischaracterises your aesthetic.
  4. Confirm rights — you must confirm you hold appropriate licences and releases for all reference material.

Training typically completes in approximately five minutes via cloud GPU infrastructure. You receive a notification when the model is active.

Step 3: Activate the model in your workspace

Once training succeeds, set the new model as your workspace active custom model. All subsequent generations in supported modules will route through it by default.

You can retain previous model versions and switch between them for A/B comparison — useful when transitioning between seasonal palettes.

Enterprise workspaces may have multiple concurrent models (e.g., diffusion vs. jewellery macro). Self-serve plans should check plan limits in pricing.

For on-model photography, personas lock casting across a session. A persona defines face, body type, and energy level — so every garment in a drop appears on the same model.

  1. Go to Personas in your workspace.
  2. Create a new persona from a reference portrait or describe casting in plain language.
  3. Save and select the persona before generating in Photo Wizard.

Personas are especially valuable for lookbooks and e-commerce hero SKUs where consistency matters more than variety.

Step 5: Generate with Photo Wizard

With your model and persona ready:

  1. Open Photo Wizard.
  2. Upload the garment reference — a flat shot, hanger photo, or supplier image works.
  3. Select your persona and brief the scene in plain English: “minimal studio, soft key light, relaxed standing pose, grey seamless.”
  4. Choose output resolution and quantity. Credit cost is shown before you confirm.
  5. Generate — review results in the session gallery. Regenerate individual frames without restarting the session.

First outputs may require one or two prompt refinements. Note what works — pose vocabulary, scene descriptors — and save session presets for your team.

Step 6: Review and publish

Before using images on PDPs or in campaigns:

  • Check garment fidelity (neckline, hem length, print placement)
  • Verify colour accuracy against your physical sample
  • Scan for artefacts (hands, background text, unintended props)
  • Confirm compliance with your AI output responsibilities

Approved images export in PNG or JPEG at PDP-ready dimensions. Add them to your DAM or upload directly to your e-commerce platform.

Common questions

Can I start before training completes?
Yes. The platform base model supports immediate generation. Switch to your custom model when training finishes — no workflow restart required.

How often should I retrain?
Retrain when your visual identity shifts materially — new season, new casting direction, or significant palette change. Minor drops often do not require retraining if your core aesthetic is stable.

What does training cost?
Custom model training consumes credits per job. See the credit table on pricing. Enterprise plans may include managed retraining.

Next steps