All insights · Sergey Kosenkov — portfolio
3D AI Workflow: From Generative Concepts to Physical 3D Printing & Virtual Production in Unreal Engine
A complete production case study on deploying generative AI models (Nano Banana Pro, Meshy.ai, Tripo3D) to eliminate modeling bottlenecks in physical manufacturing and on-set ICVFX filmmaking.
Sergey Kosenkov, Virtual Production (ICVFX) Specialist & AI Technologist · · 7 min read
Custom Avant-Garde Tableware for "Volcano" Restaurant
From Nano Banana Pro concept generation to functional 3D-printed tableware
Stage 1: Client Brief & AI Concept Design (Nano Banana Pro)
Owners of a premium Japanese cuisine restaurant approached me with a challenge: design custom, extravagant tableware for serving sushi and rolls. The core concept revolved around a massive volcano featuring a dramatic presentation—the crater needed a dedicated compartment for dry ice (creating a thick fog eruption effect), while the cascading rocky slopes provided spiral tiers for arranging sushi and rolls.
In the first phase, I leveraged Nano Banana Pro to rapidly explore visual styles and generate concept art. This allowed us to instantly iterate on the volcanic rock texture, tier proportions, and physical ergonomics before starting any 3D modeling.
AI Concept 01: Multi-tiered Volcano Platter (Nano Banana Pro)
AI Concept 02: Dry Ice Serving Concept
Preparing Orthographic References
Isolating geometry on a neutral background for 3D reconstruction in Meshy / Tripo3D
Stage 2: Orthographic Renders on a Clean Background
Once the client approved the visual concept, I needed to prepare clean orthographic references for 3D reconstruction. To achieve this, I re-generated the approved design in Nano Banana Pro on a clean, shadowless white background.
We obtained two key angles: a Top View to lock down the circular tier radii and spiral ledge proportions, and a Side View to control the overall volcano height and dry ice crater depth. The neutral background eliminated unwanted geometric artifacts during subsequent 3D AI generation.
Top View: Contour & Tier Layout
Side View: Profile & Crater Depth
3D Mesh Reconstruction & STL Export
Image-to-3D volume generation and 3D printing preparation in Meshy.ai
Stage 3: 3D Reconstruction & Print File Preparation (Meshy.ai)
The prepared 2D projections were fed into Meshy.ai using Image-to-3D mode. The algorithm generated a complete 3D volume, faithfully reconstructing the spiral rocky tiers, the cylindrical crater cavity, and a solid monolithic mesh.
Within the Meshy workspace, the model underwent mesh integrity verification to guarantee manifold topology, zero non-manifold edges, and correct face normals. The final 3D object was then exported to an engineering-grade .STL format for direct import into the 3D printer slicer.
Meshy.ai Workspace: Generated 3D mesh ready for 3D printing (STL export)
Physical Product & Business Impact
Integrating 3D-printed tableware into restaurant service & boosting average check
Stage 4: On-Site Implementation & Business ROI
The 3D printing results exceeded all expectations: the finished volcanic platter seamlessly integrated into the restaurant's premium interior and aesthetic. The functional dry ice crater produces a dense cascade of fog that flows dramatically over the tiers of fresh sushi, rolls, and sashimi.
The theatrical serving style generated immense customer excitement, with guests frequently recording videos for social media. Dishes served on the "Volcano" platter quickly became bestsellers, directly driving a noticeable rise in average check size and proving 100% ROI for the 3D AI pipeline.
3D Printed Platter with Fresh Rolls
Showcase Presentation: Dry Ice Eruption
Close-Up Detail: Rocky Slope Texture
Plated Nigiri & Sashimi Selection
3D Generative AI in Filmmaking & Unreal Engine
On-set asset generation and Art Deco banquet hall assembly under NDA
Stage 1: On-Set Virtual Environment Revisions
The second key use case for 3D generative AI is in Virtual Production (ICVFX). When filming movies and commercials against massive LED Volume walls, the virtual environment is rendered live in real time from Unreal Engine scenes. Generative AI enables creators to produce custom 3D assets in minutes directly during live film shoots.
Stage 2: Initial Unreal Engine Level Assembly (Art Deco Hall)
On a real film project (currently under NDA), the goal was to assemble a virtual Art Deco banquet hall location. The initial level in the Unreal Engine Viewport contained basic architectural geometry. During discussions, the director and production designer decided to replace the columns with a more practical design that prop makers could quickly fabricate for the physical studio set.
Initial Unreal Engine Level State: Base Art Deco Banquet Hall
Rapid 3D Mesh Generation in Tripo3D.ai
Converting prop workshop 2D sketches into production-ready 3D models in 15 seconds
Stage 3: Aligning Physical Props & Virtual Scenes
To achieve a seamless blend between the physical columns on the studio stage in front of the LED wall and their virtual extension in Unreal Engine displayed on screen, I requested an exact 2D design sketch of the octagonal light marble column from the art department.
Stage 4: Instant 3D Generation in Tripo3D.ai
The received sketch was immediately uploaded to Tripo3D.ai. In just 10–15 seconds, the algorithm converted the 2D sketch into a clean 3D mesh, preserving the octagonal profile, capital moldings, and base bevels—preventing any costly downtime on set.
Prop Workshop 2D Sketch
Generated 3D Mesh in Tripo3D.ai (~15 sec)
Unreal Engine Integration & Final Result
Replacing columns across the level and delivering the virtual set before camera roll
Stage 5: Asset Import into Unreal Engine & Final Output
The generated 3D column model was immediately imported into Unreal Engine and instantiated across all primary structural points in the banquet hall level. The Tripo3D mesh fit the scene geometry flawlessly, matching the height, scale, and proportions perfectly.
The result fully satisfied the director and production designer: the virtual columns on the LED screens became a 100% seamless extension of the physical prop columns on stage. Using 3D AI saved hours (or days) of manual 3D modeling, delivering the completed level right on schedule for camera roll.
Updated UE Level: Full Hall Overview with Columns
Camera Viewpoint: Seamless Physical/Virtual Match
Recreating Cocktail Tables for the Welcome Lounge
Image-to-3D generation of real store furniture in Tripo3D.ai
Stage 6: On-Set Furniture Matching Request
The project required setting up an adjacent Welcome Lounge with cocktail tables for greeting guests. The production designer sent a catalog screenshot of a high-top banquet table with a white spandex stretch cover purchased from an online store for physical close-up shots.
Stage 7: Instant 3D Generation in Tripo3D.ai
The product screenshot was uploaded directly into Tripo3D.ai. Within seconds, the AI generated a matching 3D asset, capturing the distinctive spandex cover contour, top fabric tension folds, and four base anchor points. This saved valuable time over complex manual Cloth Simulation, producing lightweight, optimized geometry.
E-Commerce Store Catalog Screenshot
Generated 3D Table in Tripo3D.ai (~15 sec)
Final Welcome Lounge Assembly in Unreal Engine
100% alignment between real physical props and the virtual LED environment
Stage 8: Seamless Physical & Virtual Prop Synchronization
The generated 3D table models were placed throughout the virtual Welcome Lounge in Unreal Engine, seamlessly blending into the interior design and creating an authentic event atmosphere.
Key Value of 3D Generative AI in Virtual Production: Within minutes, the crew receives specialized 3D assets that mirror physical onset props with 100% accuracy. This guarantees absolute visual continuity between practical items on set and the virtual world displayed on LED Volume screens.
Unreal Engine Viewport: Cocktail Table Placement
Camera Viewpoint: Complete Prop Matching
3D AI Generators as the New Industry Standard
Final takeaway on incorporating generative 3D tools into daily production workflows
🚀 Executive Conclusion
Generative 3D tools like Meshy.ai and Tripo3D.ai are fundamentally revolutionizing daily workflows in the 3D industry, evolving from experimental neural networks into essential production standards. They empower 3D artists, art directors, and Virtual Production supervisors to eliminate tedious modeling bottlenecks, rough blocking, and manual simulations. Whether preparing custom product designs for 3D printing in under 30 minutes or rapidly updating virtual sets in Unreal Engine during live film shoots—3D generative AI delivers 100% ROI. It frees artists to focus on creativity, dramatically accelerates project delivery, and significantly elevates final deliverable quality for clients.
⚡ Key AI Pipeline Advantages:
- Drastic Turnaround Acceleration: Saving hours to days of manual 3D modeling and complex cloth simulations.
- Seamless Virtual Production: Exact real-time alignment between physical stage props and LED Volume virtual sets.
- Production-Ready Assets: Automated geometry creation, watertight manifold verification, and PBR texturing for Unreal Engine and 3D printing.