On this page
Vrandex Studio uses artificial intelligence to help you design marketing emails. This page explains what the AI does, which models we use, what you need to check, and what you need to tell the people who receive your emails.
1. What the AI does#
- Writes and designs emails from your brief, your Brand DNA and your store data: layout, copy, subject lines, product picks.
- Creates and edits images: hero images, product scenes, backgrounds, background removal, text inside images.
- Orbi, our AI agent, watches your store, key dates and results, and prepares plans and draft emails. Orbi never sends an email or changes your store. Every draft waits in "To approve" until a person on your team approves it.
- Quality checks review designs for broken links, contrast and readability, and give a score.
2. Which models we use#
| Task | Provider | Model family |
|---|---|---|
| Text, layout, design, agent, quality checks | Anthropic | Claude |
| Image generation and editing | Google (paid Gemini API) | Gemini image models |
| Image generation and editing | via fal.ai | OpenAI GPT Image, Recraft, ByteDance Seedream, Bria |
| Hero images | Higgsfield | Soul |
| Background removal | fal.ai / Photoroom | BiRefNet, Photoroom |
We may change models to improve quality; the current list is on our Sub-processors page.
3. What you must check#
AI can be wrong. Before you approve an email, check:
- prices, discounts, codes, dates and deadlines, and that price reductions follow the 30-day lowest-price rule in your market;
- product details: names, colours, sizes, what is included, and that images show the product as it really is;
- claims: health, environmental ("eco", "sustainable"), "best", "No. 1" and similar claims must be true and allowed;
- links go to the right pages;
- logos and brands: no third-party logos or look-alike designs;
- people: no real or recognisable person appears without their written consent;
- translations and legal wording.
You are the sender of your emails and responsible for their content.
4. AI-generated images and labels#
Since 2 August 2026, EU law (AI Act, Article 50) requires AI-generated content to be identifiable.
- Machine-readable marking. Images created or edited with Vrandex Studio carry metadata (C2PA / IPTC "Digital Source Type: trained algorithmic media") and, where the model adds them, invisible watermarks. Do not strip this information.
- Visible label. If an image shows a realistic person, object, place or event that could be taken as real (for example a photorealistic model wearing your product), EU law may require that the people who see it are told it is AI-generated. This is the sender's (your) decision and responsibility. Vrandex Studio offers a small "AI-generated image" label you can switch on for your brand (Brand DNA) or for single images. It is off by default.
- Ordinary product layouts, graphic designs and promotional copy generally do not need a visible label.
5. What we never let the AI do#
- create sexual or sexualised images of anyone, or anything involving minors;
- imitate real people or celebrities without consent;
- send emails, change your store or spend money on its own.
Prompts and images that break these rules are blocked.
6. Your data and AI training#
- Our AI providers process your data only to produce your result. Anthropic and paid Google services do not train on it.
- Higgsfield may use inputs to improve its own models under its terms. Workspace owners can switch Higgsfield off in Workspace settings → "Use Higgsfield for images".
- We may use de-identified design data (layouts, structures, edits, AI-written texts with personal data removed) to improve Vrandex Studio, unless you switch it off in Workspace settings → Privacy. We never use Shopify data, your photos or logos, or third-party AI images for this. Details: Privacy Policy, section 7.