Reimagining 3D Modeling Workflows with GenAI: The Lowkey Foolish Pivot of a UX Designer.

“Stay hungry. Stay foolish.” I always liked the sound of that. But I didn’t feel its truth until I caught myself halfway through a 3D modeling sprint, trying to align fillets and tweak NURBS surfaces at 2 a.m.. Fiddling with prompts and #safetensors.

This article isn’t about perfecting design. It’s about curiosity. About reigniting old passions. About letting the pixels take a breather so I could ask myself, “What else can I build?” — and rekindle an old flame called 3D modeling.

It had been over three years since I touched 3D seriously. Not because I fell out of love, but because I got busy being someone else — a UX designer in client meetings, sprint boards, Figma files, and time zones that blended my evenings with someone else’s mornings.

And then, along came the Design Burger x Naya Studio competition: “Creative Chaos to Organized Calm.” Sounds poetic, right?

TL;DR: Still hungry to explore new tools. Still foolish enough to keep iterating. Scroll down for the key takeaways—and a few renders that made the learning feel worth it. #genai modals are too difficult and time consuming to handle for 3D Modelling workflows.

🛠️ The Modelling Journey: Nostalgia Meets Reality

My thoughts were ambitious — with #Grasshopper, I can use Parametric sliders, logic trees, clean configurations… and easily develop multiple arrangements of the cubes. It felt like building a deck of cards that is setup to crash with your own breath :)

Refined Prompt - Create a vintage-style comic book cover for the software “Grasshopper” using only the visual elements presented in the provided Grasshopper script screeh prompts & #nshot. The design should emulate the aesthetic and layout style of 1980s-90s Indian comic books, with visual inspiration from Amar Chitra Katha. Avoid any depiction of the grasshopper insect. Emphasize bold lines, primary colors, textured backgrounds in an exaggerated, retro-comic fashion. Include nostalgic flourishes like issue number, fake price tag, and an old-school publisher emblemSo I pivoted. Again and again. "Am I foolish enough to write an article about 3D Modelling being a UX Design practitioner? - Yes" But when have we drawn boundaries ? Just like the geographies, these too are fictional in my POV.

Anyway's these are #3dmodelling & #rendering #toolstack I juggled

• #Shapr3D: Browser-based, speed, but lacked the finesse.
• #Grasshopper: Cool in theory, hard to practice.
• #SolidWorks: Easy with a blueprint approach, other wise a creative bottleneck.
• #Rhino: Like bumping into your ex(tool) — nostalgic but confusing.
• #KeyShot: My old rendering buddy… until the license expired.

I wasn’t trying to get it right — I was trying to see how far I could go in the time I had.

• I stuck with Rhino, it felt like my muscle memory was no more, I had to search commands, use mouse to explore the interface, back then I used to handle rhino with keyboard commands 80% of the time ..

As the lighting setup in twin motion its taking time, i thought of bridging the gap with AI. My aim was to give it a basic render and the AI will give me professional and upscaled render. I got CHATGPT , Gemini Pro, Perplexity premium. But bouy , I failed like a Pro :-P

Prompt: Take this render of a modular concrete and bamboo desk organizer system as the fixed base image.

⚠️ Do not change or alter: Any geometry, proportions, size, or dimensions of the concrete blocks and wooden plates. Camera angle, perspective, or lighting direction and orientation of materials (matte concrete and oiled bamboo).

🎨 Enhance only by adding photorealistic accessories and environment: Place a small succulent plant with visible soil inside the tall, vertically stacked block on the left. Insert Pantone cards in the longer, double-slit wooden plate (angled slightly backward, colors visible). Add sticky notes in the small shallow tray with rectangular slits. Place pens and markers vertically into the circular holes of the wood plate (use neutral tones like black and white). Insert a smartphone in one of the backward-slanted grooves of the longer horizontal block, positioned naturally.

🌿 Lighting & Background: Enhance with soft daylight effect and gentle natural shadows for depth. Add a minimal wooden tabletop surface and a subtle neutral workspace background.

⚡️ Strict Rules: Treat the base render like a photograph; accessories must blend seamlessly. No warping, scaling, or repositioning of any existing objects. Maintain ultra-high resolution and hyper realistic quality.

I tried. I failed. And then I decided to finish it the old-fashioned way.

I went back to Rhino — carefully building the model block by block, filleting edges, splitting surfaces to expose details, manually trimming the mesh to get the precision I needed. No shortcuts.

When I finally moved the model into Twinmotion, something shifted.

This time, #AI became the teacher.

Not the kind that gives you answers — but the kind that nudges you in the right direction. It helped me understand lighting setups, scene compositions, how subtle tweaks to exposure or shadows can completely change the emotional tone of a product render.

But here's the truth no one says out loud: Don’t confuse AI with easy.

Tools like #GenAI are powerful, yes — but they demand something from you too- patience, resilience, and clarity of communication. Because AI is partially blind. It can hear what you say, but it can’t see what you mean — unless you guide it with precision.

And AI hallucinates. (Hallucination: when an AI generates false or completely made-up information because it lacks contextual understanding — especially when prompts are vague or poorly structured.)

That’s where your skill comes in to be the director, the one who can shape intent into instruction. So no, there is no magic button.

So the image you see below? It’s a render I managed to create within just a few hours — and mind you, it was my first time using Twinmotion.

Naturally, I thought: “Great start. Let’s take this to the next level. I have AI. I can do this.”Turns out — AI had other plans. Let’s just say… I got humbled. Corrected. And a bit obsessed. I underestimated the grind — and got schooled.

See the output from the big boys of AI.

#Midjourney & #VIZCOM Prompt : A photorealistic, ultra-detailed product render of a modular desk organizer set made from matte dark concrete cubes with clean fillets and removable reddish-orange cork trays. These cubes are arranged neatly on a black rubber desk mat featuring subtle neon green grid lines.

Starting from the top-left and moving right, the first unit consists of two horizontally joined concrete cubes. A black iPhone stands vertically in the center, partially hiding several horizontally stacked green markers aligned behind it.

Next to it, a single cube with front and top faces open holds a red eraser neatly nested inside a white cello tape roll. Behind this setup are several wide-capped green markers arranged upright, flanked by pastel-blue PANTONE reference cards stacked vertically at an angle.

The third cube in the layout is a vertically stacked pair supporting a cylindrical green desk lamp. The lamp emits soft white light from a round disc-like head, casting gentle illumination downward — ideal for desk environments.

To the right is a standalone cube with three vertical cork-lined slots: one holding a metallic silver pencil, another containing a green marker, and the last with an Apple Pencil in matte black finish.

Adjacent to that sits a unique concrete cube with an inset tray featuring a wide circular hole. Inside the hole is a transparent glass test tube filled with water, housing a single plant stem with large, lush green leaves, adding a biophilic element to the setup.

The last unit in the back row contains five sharpened black pencils, a vivid orange pair of scissors, and a green cutting blade — all precisely fitted into a reddish-orange cork insert with horizontal slits designed to securely hold each item. The insert is embedded into a square concrete container matching the modular design language.

In the foreground, another cube holds a stack of orange sticky notes precisely fitted into the cork insert.

The entire scene is styled with soft natural window lighting, diffused evenly across the objects to eliminate harsh shadows. Retexture all surfaces with high-resolution, professional-quality materials — enhance cork grain, concrete pitting, glass clarity, light bloom from the lamp, and micro surface reflections. Apply global illumination, accurate material bounce light, and neutral white balance to emulate cinematic product photography aesthetics. The background remains minimal and soft, ensuring full attention on the objects.

Output - #VIZCOM >> #Midjourney

These models may lack the necessary training data, or perhaps the starting image I provided wasn’t ideal. Either way, there's clearly more to uncover and learn from this.

My closest success so far came with offline AI in my drive called LM Studio (Oops it's limited to language models). For Image generation I know #ComfyUI works better, but its an ocean out there and the node connections take too much time there.

I need a web SD GUI that can do the job with a beginner friendly UI. My search bumped me to Automatic1111 and I loaded the HyperVAE.safetensors checkpoint .

I used the same prompt I'd given to #Midjourney, but the results were drastically off. That led me to dive deeper into prompt mechanics, especially understanding the role of positive and negative prompts.

Eventually, I realized my system specs couldn’t handle 4K rendering, so I had to scale down and settle for Full HD resolution.

Positive prompt : A photorealistic, ultra-detailed product render of a modular desk organizer set made from matte dark brown concrete cubes...

Negative prompt : poor anatomy, disfigured, mutated, deformed, wrong proportions, extra limbs, broken shape, inaccurate geometry, distorted, blurry, noisy, overexposed, underexposed, color bleeding, cartoonish, bad texture, plastic texture, low detail, changed shape, incorrect lighting, changed layout, changed perspective, warped materials, altered proportions, messy background, grainy render, sketch style, painting style.

These are the renders I’ve successfully refined in terms of lighting, shadows, and textures. I’m not quite at 100% yet—but I’ve learned a lot through this process. Staying hungry to explore more possibilities, and staying foolish enough to keep trying again and again.

🤝 Open to Collaborate or Just Share Notes

I’ve gathered the insights, bits that might spark curiosity or resonate with your own creative path. This is just the tip of the iceberg—there’s so much more beneath the surface.

→ #Collaboration #GenerativeDesign #PromptEngineering #CreativeTech #AIworkflow #DesignThinking #LetsConnect

If you think we could #collaborate on some hobby projects, feel free to reach out—I’d love to connect. I’m happy to dive deeper and share everything I’ve learned so far: from refining prompts and choosing the right models for different outputs, to optimizing #Twinmotion workflows and 3D modeling. I can also help with setting up a zero-cost local #LLM #ai, configuring image upscaling tools like #StableDiffusion, and leveraging virtual memory for faster, more efficient #rendering without expensive hardware upgrades. Let’s exchange ideas and push creative boundaries together.

Huge shoutout to Videolancer for the sound effects and creators on #sketchfab & #turbosquid.

🧠 Learning to Design with AI

From messy renders to confident output—what I’ve learned playing with generative design tools.

✦ Discover

Started out just curious. I’d thought about using #StableDiffusion and #Midjourney in this project to speed up things. No real plan, just testing prompts and seeing what sticks. The early results? Wild and chaotic. But oddly inspiring.

→ #DesignExploration #stayhungry

✦ Define

After enough trial and error, I noticed what kept going wrong—bad lighting, weird textures, model mismatches. So I slowed down, tracked what worked, and built a mini workflow to catch misfires before wasting render time and AI credits. It taught me how AI “thinks”—and where human instinct still leads.

→ #Promptengineering #designthinking

✦ Develop

That’s when it clicked. I set up Automatic1111 locally, loaded models like HyperVAE.safetensors, and began chaining prompts with intent. Learned how to stretch my GPU #virtualram, clean up my lighting, and stop over-complicating things. Each tweak was like solving a puzzle—with adjectives, model settings, and memory management.

→ #DIYRenderRing #ModularWorkflow #GenerativeAI #AIArt

✦ Deliver

Now I’ve built a personal pipeline to:

• Choose the right flow for the goal. Treat #genai workflow as another software, not a random chatbot.
• Refine prompts for clarity and control based on the modal and task.
• Deployed ai modals locally to generate images without huge subscription costs.
• Feed outputs into storytelling tools like Twinmotion to create videos which can act as an input for ai modals

→ #GenerativeDesign #AIinDesign #stayfoolish