At Polar Hedgehog, we work with startups that need to move fast - but without losing their identity. Every client has its own carefully developed visual language, and it’s our job to make sure that language stays consistent across websites, marketing materials, social posts, blog covers, and more.

Here’s the problem: most image-generating LLMs can give you a good single result, but struggle to stay consistent across multiple images, or to adapt to a brand’s already mature style.

That’s why we built a workflow in ChatGPT that lets us generate images with a consistent style. It adapts to the pace of startups while keeping quality high and identity intact.

Step 1: Ask ChatGPT to Capture the Style

The process starts with something simple: just ask ChatGPT to describe the style of your reference image in JSON format.

For example, you can paste in an image and write:

“Create me a JSON prompt based on the reference style.”

ChatGPT will then generate a structured description that covers every detail of the image - the material, lighting, finish, depth, background, and composition.

That JSON becomes your style blueprint for all future images.

  
    {
      "object": "3D balloon arrow",
      "style": {
        "material": "metallic foil",
        "finish": "shiny, reflective",
        "color": "silver",
        "lighting": "soft light, smooth highlights",
        "shape": "arrow pointing upward",
        "edges": "rounded, inflated seams visible",
        "depth": "pillow-like, volumetric",
        "shadows": "subtle and diffused"
      },
      "background": {
        "color": "solid bright blue",
        "texture": "flat, no gradients"
      },
      "composition": {
        "perspective": "slightly angled",
        "focus": "sharp focus on object",
        "framing": "object centered"
      }
    }
  

Step 2: Swap the Object, Keep the Style

Once you’ve got your JSON style blueprint, you don’t need to rebuild it every time. The beauty of this workflow is that you can reuse the exact same style - and just change the object you want to see.

All you need to do is ask ChatGPT something like:

“Using the same style, create a JSON prompt for a 3D balloon rubik’s cube.”

ChatGPT will take your original style description and adapt it to the new object, while keeping the material, lighting, background, and composition consistent.

  
    {
      "object": "3D balloon rubik's cube",
      "style": {
        "material": "metallic foil",
        "finish": "shiny, reflective",
        "color": "yellow, white, pink, blue, orange",
        "lighting": "soft light, smooth highlights",
        "shape": "rubik's cube",
        "edges": "rounded, inflated seams visible",
        "depth": "pillow-like, volumetric",
        "shadows": "subtle and diffused"
      },
      "background": {
        "color": "solid bright blue",
        "texture": "flat, no gradients"
      },
      "composition": {
        "perspective": "slightly angled",
        "focus": "sharp focus on object",
        "framing": "object centered"
      }
    }
  

Different objects, same visual identity. That’s how you scale consistency without reinventing the wheel every time.

Step 3: Scale Consistency

Once you’ve locked in a style, the real value comes from being able to scale it across every channel without losing quality or speed. Instead of treating each image as a one-off, this workflow lets us build a library of visuals that feel like they’re part of the same family.

With a single style definition, we can generate visuals for websites, blog covers, social media campaigns, and pitch decks.

This is where AI stops being a novelty tool and becomes a scalable design system. Consistency is no longer something we repeat manually, it becomes something we deploy strategically.

Step 4: Treat Prompts Like Code, Not Text

The final step is simple but critical: always structure your prompts as code blocks.

When you paste JSON directly into ChatGPT without formatting, it risks being treated as plain text, which can cause broken or inconsistent results. Wrapping your JSON inside triple backticks (```) signals that this is structured input the model should follow exactly.

Here’s the difference:

✅ Correct (with code block):

  
    {
      "object": "3D balloon flower",
      "style": {
        "material": "metallic foil",
        "finish": "shiny, reflective",
        "color": "purple",
        "lighting": "soft light, smooth highlights",
        "shape": "flower",
        "edges": "rounded, inflated seams visible",
        "depth": "pillow-like, volumetric",
        "shadows": "subtle and diffused"
      },
      "background": {
        "color": "solid bright blue",
        "texture": "flat, no gradients"
      },
      "composition": {
        "perspective": "slightly angled",
        "focus": "sharp focus on object",
        "framing": "object centered"
      }
    }
  

❌ Incorrect (without code block):

{ "object": "3D balloon flower", "style": { "material": "metallic foil", "finish": "shiny, reflective", "color": "purple" … } }

That small formatting step is what transforms a “nice image” into a repeatable design system.

Why This Matters

Startups don’t just need visuals that look good, they need visuals that look consistent, move fast, and stay aligned with their brand’s unique identity.

With this ChatGPT workflow, we’ve shown how a simple process

1. Asking ChatGPT to capture a style in JSON
   
   
2. Swapping objects while keeping the style locked
   
   
3. Scaling visuals across every channel
   
   
4. Formatting prompts as code for reliability
   

- turns AI image generation from a one-off experiment into a scalable design system.

At Polar, this is how we adapt cutting-edge tools to the real needs of startups:

• Fast enough to keep up with product launches and campaigns
  
  
• Consistent enough to protect the brand’s visual language
  
  
• Flexible enough to scale across websites, blogs, social, and beyond
  
  

AI isn’t about shortcuts, it’s about building workflows that give startups the speed they need while maintaining the high-quality design they expect. That balance is where Polar shines.