A gold picture frame floating in a sunset sky with ribbons of colour flowing out to a small ship model, a sound wave, a film strip and a cube that share its colours

Quick answer

To turn one reference image into a visual style brief, ask an AI model to describe the image’s colours, lighting, shapes, textures and mood, then edit that into a short list of rules (with hex colours and “always” and “never” lines) and paste it into every prompt for images, video, 3D models and voice. Check each new asset side by side with the reference and update the brief when something drifts. It takes about 20 minutes and keeps a whole project looking like it belongs together.

Key takeaways

  • One strong image beats a long description. It gives every tool the same target.
  • Let AI describe it, then you edit. Its first description is a draft, not the brief.
  • Be specific: named colours with hex codes, light direction, materials, camera angle.
  • Add “never” rules. They prevent most drift.
  • Check side by side and keep improving the brief as you go.

This guide follows on from Build a Game With Claude Fable 5.2 and Higgsfield MCP: What One Chat Really Did. In that build, WorldofAI started with a single piece of key art for a neon racing game, then asked Claude to break down what the game should copy from it: the cyan and magenta neon, the wet reflections and the ship’s shape. That one picture guided the code, the 3D ship and the trailer.

📘 Main article: Read how the whole game came together, from key art to trailer, in Build a Game With Claude Fable 5.2 and Higgsfield MCP: What One Chat Really Did.

This guide answers one question: how do you turn one reference image into a written style brief that keeps every AI-made asset consistent?

Why use one reference image as your style guide?

Because AI tools interpret words differently, but an image is the same for everyone. Ask three tools for “a modern neon look” and you will get three different looks. Give them the same reference and the same written rules, and the results start to match.

This matters well beyond games. A product launch, an online course or a social campaign all use images, short videos and sometimes voice. If they look unrelated, the whole thing feels less professional.

What do you need?

  • One reference image you are allowed to use: one you made, generated or licensed.
  • An AI model that can see images, such as Claude, ChatGPT or Gemini.
  • A document or note to keep the brief in.
  • 20 minutes for the first version.

⚠️ Warning: Do not use someone else’s artwork or a film still as your reference for anything you plan to publish or sell. Make or license your own. Our checklist What to Check Before You Publish an AI Image: Text, Likeness and Usage Rights covers usage rights.

Step 1: Choose or make the reference image

Pick the one image that best captures how the finished project should feel. Good references have:

  • a clear main subject;
  • a limited colour palette, usually three to five main colours;
  • distinctive lighting, such as neon at night or soft daylight;
  • enough detail to show materials and textures.

If you do not have one yet, generate it first, as WorldofAI did. Spend your effort here: every later asset will copy this image’s strengths and its faults.

💡 Tip: If you need to blend a logo, a product shot and a mood into one reference, see How to Combine Multiple Reference Images in One AI Image.

Step 2: Ask AI to describe it

Upload the image and use this prompt:

“Describe this image as a style reference for a designer. List: the main colours (with approximate hex codes), lighting and time of day, materials and textures, shapes and proportions, camera angle and framing, level of detail, and overall mood in three words. Then list five things that make it look distinctive.”

You will get a long, useful description. Do not paste it everywhere as it is. It will contain guesses and repetition.

Step 3: Turn the description into a style brief

Edit the description down to about ten short lines. Keep only what you can see and what you want repeated. Check the hex codes with a free colour-picker tool and correct them if needed.

Then add two short lists:

  • Always: three to five things every asset must include, for example “cyan engine glow”, “wet reflective surfaces”, “night setting”.
  • Never: three to five things that would break the look, for example “daylight”, “pastel colours”, “cartoon outlines”, “text in the image”.

Step 4: Add rules for each type of asset

Different tools need slightly different instructions, so add one line for each asset type you plan to make.

AssetWhat to add to the briefExample line
ImagesAspect ratio and framing“16:9, subject slightly off-centre, low camera angle”
VideoMotion and pacing“Slow camera push, fast streaks of light, no hard cuts”
3D modelsClean views first, materials“Generate clean side, front and top views on a plain background before modelling”
Voice and soundTone that matches the look“Energetic, clear announcer; synth sounds, no acoustic instruments”
Web pages or game screensColours and fonts for interface“Dark background, cyan buttons, magenta highlights”

The 3D line comes straight from the WorldofAI build: the rainy key art was too busy to convert directly, so clean views of the ship were drawn first. That is a good habit for any reference full of atmosphere.

Step 5: Use the brief in every prompt

Paste the brief at the top of every prompt, then add the specific request underneath. For example:

“Use this style brief: [brief]. Now create: a square image of the racing ship parked in a garage, same colours and lighting.”

If you are working with an AI agent that calls other tools, like Claude with an MCP connector, ask it once: “Save this style brief and apply it to every asset you create in this project.” Then check that it does. If you have not set up a connector yet, our guide How to Add a Custom MCP Connector to Claude Safely walks you through it.

For video, pair the brief with a well-structured prompt; our guide How to Write an AI Video Prompt That Models Actually Follow shows how.

Step 6: Check new assets against the reference

Put each new asset next to the reference and score it quickly. A simple checklist works:

CheckYes / No
Same main colours?—
Same lighting and time of day?—
Same materials and textures?—
Follows every “always” rule?—
Breaks no “never” rule?—

If an asset fails, do not just regenerate and hope. Find the rule that was missing or vague, add it to the brief, then regenerate. Over a project, the brief gets sharper and the assets get closer.

What should the brief look like?

Here is a simple template you can copy:

SectionYour notes
Project and mood (3 words)e.g. “HYPERLUME: fast, neon, rain-soaked”
Main colours (hex)e.g. cyan #00E5FF, magenta #FF2BD6, deep navy #0A0F2C
Lightinge.g. night, neon glow from below, strong reflections
Materials and texturese.g. glossy metal, wet tarmac, glass
Shapes and proportionse.g. long, low, sharp angles
Camera and framinge.g. low angle, motion blur on background
Always3–5 rules
Never3–5 rules
Per-asset linesimages, video, 3D, voice, interface

What mistakes should you avoid?

  • Pasting the AI’s first description unedited. It is too long and some of it will be wrong.
  • Vague colour words like “blue” or “warm”. Use names and hex codes.
  • Too many rules. Ten clear lines work better than a page.
  • Changing the reference halfway through. If you must, update the brief and recheck older assets.
  • Forgetting non-image assets. Voice, music and interface colours are part of the look too.

Frequently asked questions

What is a visual style brief?

A short written description of how a project should look, covering colours, lighting, materials, framing and mood, plus rules about what to always and never include.

Why not just upload the reference image every time?

Do both where you can. Some tools cannot take image references, and written rules stop the model copying the wrong details.

How long should the brief be?

Around ten lines plus short “always” and “never” lists. Longer briefs tend to be followed less closely.

Can I use this for brand content, not games?

Yes. It works for any set of assets that should feel connected, such as a launch campaign, course materials or social posts.

Which AI is best at describing images?

Most current models with vision, including Claude, ChatGPT and Gemini, do this well. Always check the colours and details yourself.

What if the AI keeps drifting from the style?

Find the missing or vague rule, make it more specific, and add a “never” line for the unwanted change.

The bottom line

A single reference image is the cheapest way to make AI tools agree with each other. Describe it with AI, edit the result into a short brief with clear colours and rules, and paste it into every prompt. That is how one key-art image could steer the code, the ship and the trailer in WorldofAI’s game build, and it works just as well for your next campaign.

For the full build that inspired this guide, go back to Build a Game With Claude Fable 5.2 and Higgsfield MCP: What One Chat Really Did.

author avatar
Garry Knight
I'm Garry Knight, the person behind Prodify Digital. I write about email list building, email marketing, SEO, AI search and the tools that connect them. My aim is to make online marketing easier to understand, so creators and small business owners can make informed decisions about building an audience and keeping people engaged. Here you'll find straightforward guides and product reviews that explain what something does, where it fits and which limitations matter. The focus is on clear explanations and useful next steps—not hype, shortcuts or promises of easy earnings.

Leave a Reply