Thoughts From a Dev

Category: Uncategorized

  • AI Image Generation has just Leveled Up

    🎨 ChatGPT Just Got a Vision Upgrade: Say Hello to On-Demand Image Creation

    You asked for it. OpenAI delivered. ChatGPT can now generate, edit, and remix images—without needing a third-party design tool or hours in Photoshop. Whether you’re building a brand, creating marketing material, or just vibing on creative chaos, it’s now as simple as typing what you want.

    🔧 What Can It Do?

    • Generate images from text prompts: Describe a scene—get the image. Hyper-realistic photos, logos, vintage posters, cinematic vibes—you name it.
    • Edit existing images: Want to add glitter text to your casino chips photo? Done. Remove background junk? Say the word.
    • Style control: From 60s psychedelic to cyberpunk noir to clean, ADA-compliant branding. You get what you ask for.
    • Photorealism or stylization: Whether you want studio-grade lighting or sketchbook-style concept art, ChatGPT can do both.

    💡 Real-World Uses

    • Marketers & Agencies: Create campaign visuals in seconds.
    • Web Devs & App Designers: Auto-generate assets for mockups or production.
    • Educators & Content Creators: Bring ideas to life visually, without needing a designer.
    • Side Hustlers: Product shots, logos, social ads—you’re no longer bottlenecked by Canva.

    1. Core Structure of a Killer Image Prompt

    Break prompts into parts:

    [Subject] + [Setting/Scene] + [Style] + [Lighting] + [Camera/Composition] + [Mood/Emotion] + [Post-Processing]

    Think of it like giving direction to an AI creative team. The more detail, the better the result.

    🧠 2. What You Can Ask For

    ✅ SUBJECTS

    • A woman with lavender hair sipping coffee
    • A 60s psychedelic mushroom poster
    • A cyberpunk alley with neon signs
    • A photorealistic hamburger with melting cheese

    Use modifiers: age, ethnicity, outfit, mood, pose, facial expression

    ✅ SCENES

    • Inside a cozy cabin during a snowstorm
    • A futuristic Tokyo street at night
    • A retro American diner at sunrise
    • A pirate ship in a stormy sea

    ✅ STYLES

    • Photorealistic
    • Hyper-realistic
    • Comic book / Ink drawing
    • 60s/70s Psychedelic poster
    • Watercolor / Oil painting
    • Concept art / Matte painting
    • Line art / Sketch
    • Low poly / Pixel art
    • Vaporwave / Cyberpunk / Steampunk
    • Brutalist / Minimalist / Maximalist

    ✅ LIGHTING

    • Soft diffused light
    • Cinematic lighting
    • Backlit silhouette
    • Neon glow
    • Golden hour / Sunset light
    • Studio lighting with shadows
    • Rim lighting

    ✅ CAMERA COMPOSITION

    • Shot on 35mm film
    • Shallow depth of field
    • Close-up / Macro shot
    • Wide-angle / Bird’s-eye view / Low angle
    • Over-the-shoulder
    • Bokeh background
    • Rule of thirds / Symmetry / Centered

    ✅ MOOD / VIBES

    • Introspective
    • Joyful
    • Trippy
    • Mysterious
    • Surreal
    • Cozy
    • Tense / High-stakes

    ✅ POST-PROCESSING / FINISHES

    • Film grain
    • High contrast
    • Color graded like Blade Runner
    • Washed-out tones
    • Glitch effect / VHS
    • Embossed on kraft paper
    • ADA-compliant text overlay
    • Vintage print texture

    🛠️ 3. Functional Prompts (Logos, Marketing, Product)

    • A clean logo embossed on recycled paper, realistic lighting
    • A “SALE” banner with glitter text on a white background
    • A modern app UI mockup, dark mode, minimal layout
    • A luxury skincare product on marble with soft shadows

    🎨 4. Design-Smart Extras You Can Request

    • “Make the font ADA compliant”
    • “Add depth of field and lens blur”
    • “Use bold serif typography”
    • “Give it a handmade, silkscreen print vibe”
    • “Add sparkles around the title”
    • “Use pastel tones and soft gradients”
    • “Make it print-ready with high resolution”

    🧪 5. Advanced Techniques

    • Image editing: Upload an image and say:
      “Add neon lights in the background”
      “Replace the text with ‘Summer Sale’ in the same style”
    • Variations:
      “Give me 3 variations of this with different lighting”
      “Same poster, but with a cat instead of a dog”
    • Brand matching:
      “Design a banner using the colors and style of the Apple website”

    🚀 6. Prompt Examples You Can Steal

    “A neon-lit sushi bar in a rainy alley, cyberpunk Tokyo, cinematic lighting, photorealistic, 35mm film grain, dramatic shadows, puddle reflections.”

    “A 60s psychedelic mushroom poster with the text ‘Great Day’, bold warped typography, vibrant warm tones, thick outlines, vintage screen-print style.”

    “Photorealistic studio shot of poker chips, dice, and cards on a green felt table, shallow depth of field, with gold glitter text saying ‘Win The Game’, ADA-compliant font.”

  • CEOs that love AI are out of touch

    CEOs that love AI are out of touch

    Chris Sacca recently said on The Time Ferriss Show that developers are useless because you can “Just tell AI to build me an app” and it will just work.

    Like some magical fariy that comes and grants wishes…

    Bullshit….

    Chris may have found out how financial systems work because he has way more money than I do, but one thing he doesn’t know… AI..

    Most people don’t actually. YouTube is filled with last months “Crypto Bros” now telling you to “Make an AI agency” and you can “Make millions of dollars, just like them”…. also more Bullshit…

    This is not a criticism on what AI can do, this is a critical observation on people that are over hyping a technology all for sensationalism and to get your YouTube clicks.

    What AI Can Do

    A lot actually.

    Smoky scent lingers,
    Sweet tang dances on the tongue,
    Bones sing of delight.

    A BBQ Rib Hiku that ChatGPT wrote for this.

    Its also responsible for the featured image of this post and many other things. Its great at language as well.

    It can be used for code, I use it all the time. But its no better than a JR dev that has memorized the syntax but doesn’t understand the real world context of software architecture and application flow.

    Its not a magic wand

    If you want AI to build an app, you need to take the time to think through everything. I spent 3 months one time going through every use case, all the workflows, etc. Building all the ideas and state flow diagrams, charts, etc. Just like you cant ask a dev on your team to “Build the whole project on you own” you can ask AI that either. Context windows are too small. I tried to have ai build a landing page, that worked great, until I asked to to change the theme of the page, styles only, but instead, it decided that the navigation shouldn’t exist any more. Then when you ask it to restore the nav, it will but now your footer is gone.

    If you want AI to build anything, you need to know how to be specific about what you ask for, with all the details, and conditions. You don’t say “build a login” you say

    “We are making a login page with full database communications. the username is an email address that should be valid, the password needs to be min 6 chars with letters, numbers, and special characters. We need to give the user 5 attempts. if the auth fails, report the error. use modern styles that are mobile responsive. We should also make sure that the form is fully accessible. “

    But that’s it, just focus on one part at a time, then move on to the next part. you do this over and over for days because of all the bugs it generates.

    Will it happen one day

    Yes, but that day is not in 2025… even with how fast it’s accelerating and o1 and deepseek’s reasoning. It’s still no where near as good as a dev with experience.