Gilad Ostrover - AI Technical Artist

Post-Apocalyptic Theme Park

Maya Depth Maps + ComfyUI + Immersive VR Content

Maya + V-Ray ComfyUI Custom LoRA 360° Panoramic VR Ready

From 3D Architecture to AI-Generated Worlds

A haunting vision of an abandoned amusement park, where rusted machinery and weathered signage tell stories of forgotten joy. This scene demonstrates the power of using architectural depth maps to create emotionally rich environments that transcend their original purpose.

6144×3072
Resolution
360°
Full Coverage
VR Ready
Equirectangular
Depth Map
Maya Z-Depth Map
Maya Scene
Original Maya Scene

Enchanted Forest Portal

Mystical Woodland Realm with Ethereal Lighting

Maya ComfyUI ControlNet Custom Training

Magical Realism

Originally modeled as construction building blocks, this depth map was transformed into a mystical woodland realm. Purple-hued mushrooms and ethereal lighting showcase how the same geometric foundation can support entirely different creative visions.

Depth Map
Maya Z-Depth Map
Maya Scene
Original Maya Scene

360° Panoramic Environment

Advanced AI-Generated Immersive Space

Maya + V-Ray Stable Diffusion Depth Conditioning

Immersive World Building

A complete workflow demonstrating the transformation of traditional 3D depth maps into fully immersive 360° environments. This showcases the seamless integration of Maya modeling with cutting-edge AI generation techniques.

Depth Map
Maya Z-Depth Map
Maya Scene
Original Maya Scene
Faldourai Robot

Custom LoRA Training

Faldourai: Vintage Robot Character Model

Kohya_ss Dataset Curation Local Training Character Generation

From Dataset to Production-Ready Character

A complete custom LoRA training project from curated dataset to production-ready character generation. This model demonstrates advanced training techniques, creative applications, and innovative problem-solving when facing technical limitations.

60+
Curated Images
3
Training Iterations
Kohya_ss
Local Training

Training Pipeline

Dataset
Dataset Curation - 60+ Images
Config
Kohya_ss Configuration

Generated Content Gallery

Generation 1
Generation 2
Generation 3
Generation 4
Generation 5
Generation 6
Generation 7
Generation 8
Generation 9
Generation 10
Generation 11
Generation 12
Generation 13
Generation 14
Generation 15

Animated Content

Flamingo Pool Float Series
Summer Vibes Animation
Vehicle Form Exploration
Mobile Robot Variant

Case Study: Breaking the Pose Limitation

The Challenge: The trained LoRA model could only generate the robot in standing poses - the dataset contained no examples of bent joints, sitting positions, or dynamic poses. Creating a bowing animation required a creative technical solution.

The Solution: A hybrid pipeline combining multiple technologies:

  1. 3D Reconstruction: Used Hunyuan3D-DIT-v2 in ComfyUI to generate base 3D model from LoRA outputs
  2. Topology Refinement: Re-topologized mesh with Quad Remesher for clean geometry
  3. Format Conversion: Converted GLB to OBJ in Blender for Maya compatibility
  4. Rigging & Posing: Manually segmented mesh into joints in Maya, posed for bow animation
  5. Depth Map Rendering: Generated Z-depth maps with V-Ray for ControlNet guidance
  6. Image-to-Video: Combined depth control with LoRA to generate final animation
Bow Process
Maya Poses & Depth Maps
Final Bow Animation

This project demonstrates the ability to combine AI tools (LoRA training, image generation, image-to-video) with traditional 3D workflows (Maya, Blender, V-Ray) to overcome limitations and achieve creative goals.

Multi-Image Meshup

Multi-Image Meshup Workflow

ComfyUI IPAdapter Pipeline for Creative Image Blending

ComfyUI IPAdapter Weighted Embeddings SDXL / SD 1.5

Flexible Blending Pipeline

A flexible ComfyUI pipeline for blending 2-5 source images into cohesive, creative compositions. Each result demonstrates precise control over visual influence through weighted IPAdapter embeddings, creating photorealistic hybrids that maintain artistic intent while combining distinct visual elements.

Creative Results

Meshup 1
Turquoise Scooter Mashup
Meshup 2
Rainbow Gamepad Mashup
Meshup 3
Green Turquoise Vehicle
Meshup 4
Retro Red Vehicle
Meshup 5
Red Bus Winter Scene
Meshup 6
Citrus Fruit Mashup

ComfyUI Workflow Architecture

Workflow
Complete Workflow Diagram

Technical Approach

This modular workflow system features specialized pipelines optimized for different content types while maintaining a consistent blending methodology:

  • Objects & Products: Optimized for mechanical forms, toys, and product design (SD 1.5 / SDXL)
  • Characters & Portraits: Specialized for facial features and human subjects (SD 1.5 / SDXL)
  • Flexible Source Control: Each generation uses 2-5 weighted source images based on creative requirements
  • Precision Weighting: Individual IPAdapter strength control enables fine-tuned influence from each source

The workflow demonstrates both technical versatility and creative problem-solving, adapting the pipeline architecture to match the specific visual demands of each project while delivering cohesive, photorealistic results.

Let's Connect

Interested in collaborating or want to discuss AI workflows and production pipelines?

Contact me at contact@gilad.studio

Open to full-time positions, freelance projects, and technical consultations.