Make it Pop #11 - Brand guidelines with AI instructions (and as API?!)

I’ve been going down a rabbit hole lately.

I’m currently building a demo where I want to generate fully on-brand assets: images, copy, layouts, based strictly on a company’s brand guidelines. The goal was a simple "lift and shift": upload the traditional PDF guideline into an AI tool, prompt it, and get a perfect assets.

It didn't work. ofc…

When I uploaded the glossy, beautiful brand PDFs we all know, the images coming out were... wrong. The vibe was off.

I had this "oh man…" moment: I am trying to give human instructions to AI.

Traditional brand books assume shared context, cultural fluency, and intuition. If you tell a human designer to make it "feel premium," they use their experience to choose matte textures and serif fonts. But AI has no intuition. It doesn't know what "premium" feels like; it only knows what "premium" calculates to.

If we treat AI like a creative hire who will just "figure it out," we are setting ourselves up for brand dilution at scale, especially now when all tech bros want to automate creativity at scale (hint hint you need a creative to set the creative direction).

We need to stop dumping PDFs into context windows and hoping for the best. We need to translate our "Vibe" into "Math." We need to move from inspiration (adjectives) to calibration (parameters).

Here is the framework I developed to solve this to make your brand guidelines AI-ready.

The AI-ready brand system template

Most guidelines explain "why." AI guidelines must explain "what" and "how much". Use this structure to turn your PDF into a machine-readable instruction set.

SECTION 1: Visual Physics ( "virtual camera")

Before you would just put bunch of images in your pdf brand guideline because the human designer could interpret it, but for AI you need to be specific! I models default to glossy hyper-realism if you don't constrain them. Literally go into your brand guidelines and add what you would use in your prompts when generating images.

• Camera & lens simulation: Don't just say "professional photography." Specify the gear.

  • Example: "Sony A7R V with a 50mm prime lens."

• Depth of field: Define the focus.

  • Example: "f/1.8. Razor-sharp focus on the subject; creamy, soft bokeh background."

• Lighting architecture: Replace mood words with lighting setups.

  • Example: "Soft, large-source window light from the left (45 degrees). Fill ratio 2:1. NO harsh mid-day sun shadows. NO on-camera flash look."

• Color grading: Define the post-processing.

  • Example: "Low saturation greens, warm highlights, high contrast blacks (Kodak Portra 400 aesthetic)."

• Whatever else defines your brand, add it to your guidelines as if you were writing prompts/prompt instructions.

SECTION 2: Asset Anchors (logo & palette)

• Palette (the HEX police):

  • Rule: "Primary: #0052FF. Secondary: #F2F4F7. Bias all background generation toward the Secondary hex code (60% coverage). Use Primary only for focal points (10% coverage)."

• Logo composition strategy:

  • Rule: “Compose the logo with a clean, high-contrast negative space in the top-left quadrant (approx. 20% of the frame). This space must be free of texture to allow for programmatic overlay of the vector file."

SECTION 3: Syntax engineer (voice & tone)

• Sentence physics:

  • Rule: "Max 20 words per sentence. Paragraphs must not exceed 3 lines."

• "Ban list" (negative constraints):

  • Rule: "Strictly forbidden words: 'Unlock,' 'Delve,' 'Tapestry,' 'Game-changer,' 'Synergy.' If the prompt implies these, replace with 'Build,' 'Create,' or 'Work'."

• Tone dialing:

  • Rule: "If the audience is 'User', use active voice (You can...). If the audience is 'Enterprise', use benefit-led voice (The platform enables...)."

SECTION 4: Guardrails (safety & bias)

• Diversity & inclusion:

  • Rule: "Scenes must feature diverse subjects in active, leadership roles. Avoid stereotypes." (you can probably get more specific here)

• Visual bans:

  • Rule: "No 3D cartoon renders. No neon/cyberpunk lighting. No text overlays generated by the model."

In a nutshell, you need to add prompt instructions into the pdf brand guideline instead of just giving reference images and hope the model interprets the lighting, camera type etc.

🔮 The future: Brand guidelines as an API

This experiment made me realize that the static PDF is actually inefficient. If you upload a 50-page PDF to a model, you are burning through tokens and confusing the context window with irrelevant info. This makes generations slower and less accurate.

I predict we are entering an era where brand guidelines won't be a file you read; they will be an API you call.

We are already seeing the prototype of this with tools like Midjourney’s Style Reference (--sref). When you use --sref, you aren't describing the vibe; you are passing the model a data file and saying, "Read the math in this image. Copy these pixel values."

That is a Brand API in its simplest form.

Imagine the near future: You won't upload a PDF to your AI tools. You will connect Canva, Figma, or Sora to your company's brand.json endpoint.

1. You prompt: "Make a launch post for Q1."

2. API intervenes: It scans your query and fetches only the relevant rules from your Brand API (e.g., "Q1 Campaign Lighting" and "Launch Voice"). This saves context window space and prevents the model from getting confused by irrelevant rules for the specific prompt.

3. Audit & compliance: Before you ever see the output, an agent (with Gemini 3) critiques it against the brand rules. If it doesn't comply (e.g., wrong hex code), it rejects the output and restarts the generation.

   (Sometimes models hallucinate so you need something like Gemini 3 to critique the outputs - I have done this in real life, and it works like magic!)

But until that API infrastructure is standard, here is what you can do today:

1. Rewrite your PDF with "AI Instructions"

2. Go to Google AI Studio.

3. Test your new guidelines with Nano Banana Pro (which actually accepts PDF input at the API level).

Good luck! Let me know what tweaks you had to make to make this work for yourself. This is probably gonna take a bunch of errors and trial. Btw, I vibe-coded an app in AI studio, fed it with brand guidelines and the app generated images in the style I instructed in the brand guideline. In my next post, I will show step by step (I just need to pick a mock brand, cuz I can’t use the brand I was testing this for originally).

📰 AI creative news updates 20-27th January 2026

• Higgsfield launched AI Influencer Studio, a tool that lets you build a custom AI character with video-game-style controls and generate consistent, viral-ready content. Users can earn income by connecting their accounts to Higgsfield Earn.

• Remotion + Claude Code can now generate full animated videos with transitions from prompts, exporting to MP4 without traditional editing tools. No After Effects or Premiere Pro, dramatically lowering the barrier to high-quality video creation.

• LTX Studio’s Audio-to-Video transforms music, voice, or effects, into a synchronized video sequence. By using the audio to drive visual motion and pacing, it allows creators to generate polished videos without the need for manual editing.

• Black Forest Labs released FLUX.2 [klein], a family of super-fast image models that unify generation + editing with sub-second output & run on consumer GPUs. This means near real-time generating & editing of images instantly on everyday hardware.

• Krea Realtime lets you see AI-generated images update instantly as you type, move shapes, or use reference images, with adjustable influence and live canvas tools.

• Qwen3-TTS is a new open-source family of advanced text-to-speech models that support voice cloning, voice design, streaming speech, and multilingual generation across many languages.

• Adobe rolled out new AI-powered video editing tools in Premiere (like AI Object Masking and Firefly Boards integration) and major motion design upgrades in After Effects to speed up effects, masking, 3D, vector and animation workflows.

🏆 AI creative competitions worth joining

If you’ve got a video or concept brewing, these competitions are open right now, and they’re giving real prizes + visibility to your creative AI work

On a personal note, I am currently brainstorming as to what to do with my life - career. I am happy where I am today, but I need a new challenge to grow! So stay tuned!

Khulan