Can I keep several brands separate?

Yes. Create one project per brand. Each project has its own Brand DNA, logo and images, so a client's look never leaks into another's.

Can I turn Brand DNA off for a single image?

Yes. The Brand DNA switch sits in the prompt bar, so you can switch it off for one generation, for example to try a completely different look, and back on for the next.

What if BrandGen reads my brand wrong?

Edit it. The Brand DNA page is an editor: fix a name, a colour or a rule, save, and every image after that uses the corrected version. We corrected one misread name in this test.

How many products can I attach to one prompt?

One. Each prompt takes one attached image, which you pick from up to five saved in the project. If you ask for three products but attach only one, the model may invent the other two, so show one product per image or combine separate images afterwards.

How do you keep AI-generated images on brand?

Yash SoniFounder of BrandGen

October 19, 20268 min readLast updated Oct 19, 2026

To keep AI-generated images on brand, give the AI a fixed brand spec that it applies to every image, keep your prompts about the subject rather than the style, attach your real logo and products, and check each image before it goes out. Without a spec, every prompt starts from the model's own default look.

Below we ran the same three prompts twice for one brand, once with its Brand DNA off and once with it on, so you can see exactly what changes.

Key takeaways

Off-brand images are the default, not a bug: each prompt starts from the model's own idea of what looks good.
A brand spec saved once beats describing your style in every prompt.
Prompts should say what is in the image; the spec should decide how it looks.
Give the AI your real logo and attach your real product, or the model invents them.
Check people, products and every word before posting. One quick refine fixes most problems.

Why do AI images drift off brand?

Because nothing tells the model what your brand looks like. Each prompt is interpreted from scratch, so the model fills every gap (light, colour, type, setting) with its own defaults, and those defaults change from one image to the next.

You can see it in the left-hand images on this page. Asked for a fashion post, the model produced a sunny Italian street with a serif headline. Asked for a launch post, a seaside rooftop with brush lettering. Asked for a pop-up poster, a New York shopfront. Each is fine on its own, and none of them look like the same brand.

It gets worse with details the prompt does not mention. With no brand context, our pop-up poster came back with a made-up shop name, a made-up street address, and "Grand Opening" and "Opening Soon" on the same image.

What are the ways to keep AI images on brand?

There are four common approaches, and they trade effort for consistency.

1. Describe the style in every prompt. Free and works anywhere, but long prompts drift: "bold, colourful, modern" means something different on every run, and nobody on the team writes it the same way twice.

Got any questions left?

Keep every AI image on brand

Turn on Brand DNA and every image you generate starts from your colours, type and style.

Try BrandGen free

Keep reading

What Is Brand DNA? Meaning and Key ElementsBrand DNA is the palette, typography, visual style and tone that make a brand recognizable. Learn what goes into it and how it differs from a style guide.7 min read

What does the same prompt look like with and without Brand DNA?

We set up an invented clothing label, velle: a colour-blocked, Gen Z look with cobalt, hot pink and yellow sets, flash lighting, a huge lowercase wordmark and small monospace captions.

Input. One past post as the brand image, the logo, and one product photo.

BrandGen read the post into Brand DNA in about a minute. It picked up the colour-blocked sets, the arches and cubes, the pill-shaped badges, the lowercase monospace captions, and a rule to avoid muted palettes. It also misread the half-hidden wordmark as "verde", which we corrected in the editor before generating.

Real screenshot of the Brand DNA page after our one correction.

Then we ran three everyday prompts twice each: once with Brand DNA switched off, once with it on. None of the prompts mention colours, fonts or style.

Pair 1, product in use. Prompt: "Instagram post: a model wearing the attached cardigan, sitting on the hood of a vintage car, laughing with her head tilted back. Headline "Lime Time". Small line "drop 08 / out friday"." (Shown at the top of this post.) Off, it is a sunny street with a serif headline. On, the scene moves onto a cobalt and pink set, the headline becomes the brand's huge lowercase type, the caption turns monospace, and the logo appears top right.

Pair 2, launch post. Prompt: "Instagram post announcing our summer drop: three friends standing side by side in different outfits from the collection, arms linked, mid-laugh, full bodies visible. Headline "The Summer Drop". Small line "out friday"."

Pair 3, event poster. Prompt: "Poster for our pop-up shop opening: a model walking out of a small shop front holding two shopping bags, full body visible. Headline "Pop-up Shop". Line "saturday, 12pm"."

In our testing, all three Brand DNA versions kept the palette, the set style and the type system, and placed the real logo or branded bags without being told what they look like. The Brand DNA off versions only ever used the words we gave them, plus whatever the model invented.

Cookie preferences

How do you keep AI-generated images on brand?

Key takeaways

Why do AI images drift off brand?

What are the ways to keep AI images on brand?

Got any questions left?

Keep reading

What does the same prompt look like with and without Brand DNA?

What should your brand image include?

How do you write prompts that work with your brand, not against it?

How do you check an image is on brand before you post it?

Try it