Visual Explorations
A collection of experimental AI-assisted workflows combining 3D, cinematic rendering, realtime tools and custom pipeline development.
Alien 3d & AI Demo
This project explores a hybrid workflow combining 3D sculpting, AI-assisted rendering, and iterative visual development.
The alien head was initially sculpted in Nomad Sculpt before being processed through multiple AI-assisted rendering and texturing passes using Nano Banana to explore different surface details, materials, and visual directions.
From there, I expanded the design into a full character sheet and used the established visual language to generate a broader set of cinematic images and animated sequences.
The project focused heavily on:
rapid visual iteration
character consistency
stylized rendering workflows
combining traditional 3D foundations with modern generative AI tools
Tools used:
Nomad Sculpt, Nano Banana, ComfyUI, Photoshop
klein & fine Inpainter
An experimental inpainting workflow focused on fast visual iteration directly inside ComfyUI.
The project explored how rough paintovers, masks, and simple compositional guides can be used to steer AI-generated environments while still preserving overall lighting, depth, and scene coherence.
To improve iteration speed and reduce dependency on external APIs, Flux Klein was integrated locally for lightweight inpainting and rapid scene exploration.
The workflow was primarily designed around:
fast environment ideation
composition blocking
AI-assisted paintovers
iterative scene refinement
interactive creative workflows inside ComfyUI
Instead of relying purely on prompting, the system allows visual guidance through direct image manipulation, making the process feel closer to traditional concept art and matte painting workflows.
Tools used:
ComfyUI, Flux Klein, Photoshop
Gaussian Splat cinematic experiment
An experimental cinematic workflow combining Gaussian splats, traditional 3D scene assembly, AI-assisted refinement, and animated post-processing.
The environment foundation was created using a Gaussian splat captured with Polycam, allowing for fast scene creation based on real-world spatial data and lighting information. A separate robot asset sourced from Blender Market was then integrated and positioned inside the scene using Blender.
After establishing the core composition and camera framing, the render was brought into ComfyUI for additional refinement passes, where environmental detail, atmosphere, and surface complexity were enhanced through AI-assisted rerendering workflows.
Additional relighting and fire effects were added to further push the cinematic mood and visual storytelling before the final sequence was animated in Kling.
The project explored:
hybrid Gaussian splat + AI workflows
fast cinematic environment assembly
AI-assisted scene refinement
relighting and atmospheric enhancement
combining real-world scans with stylized rendering pipelines
Tools used:
Polycam, Blender, ComfyUI, Kling, Photoshop
3D to Ai render experiment
A hybrid concept-to-cinematic workflow combining sketching, AI-assisted rendering, lightweight 3D production, and animated post-processing.
The project started with a rough robot concept sketched in Procreate, which was then expanded into a more realistic character sheet using Nano Banana to explore proportions, materials, and overall visual direction.
To establish a controllable production foundation, the character was converted into a rough 3D asset using Meshy AI and placed into a simple environment built in Blender for initial composition, camera framing, and scene structure.
The rendered scene was then brought back into ComfyUI, where additional refinement passes were used to enhance the robot design, add environmental foliage, and further develop the atmosphere and visual complexity of the shot.
Finally, the sequence was stylized using a painterly Nano Banana pass before being animated in Kling.
The workflow focused heavily on:
rapid concept-to-scene iteration
hybrid AI + 3D production pipelines
cinematic environment development
controllable composition workflows
combining rough foundational assets with AI-driven refinement
Tools used:
Procreate, Nano Banana, Meshy AI, Blender, ComfyUI, Kling, Photoshop
Motion Echo Editorial
An experimental AI fashion editorial exploring motion echoes, latent-space manipulation, and stylized temporal distortion within cinematic urban environments.
The project was built around a custom ComfyUI workflow combining multiple AI-assisted image processing techniques with bespoke node development.
To create the signature visual repetition and stretched motion trails, I developed two custom nodes together with ChatGPT:
a latent displacement node capable of shifting latent information along the X and Y axis
a custom echo node designed to generate layered temporal repetition and motion distortion effects
Rather than relying purely on prompting, the workflow focused on directly manipulating image structure and latent-space behavior to achieve more controllable and visually distinct results.
The project explored:
experimental latent-space manipulation
AI-assisted fashion editorials
custom ComfyUI workflow development
motion echo effects
cinematic color contrast and visual rhythm
iterative AI-driven art direction
Tools used:
ComfyUI, custom nodes, Z-image, ChatGPT-assisted prototyping
Dark Fantasy Character Exploration
A cinematic dark fantasy character exploration developed entirely inside ComfyUI using a hybrid AI-assisted workflow focused on iterative character and environment development.
The project started with rapid visual exploration using Z-Image Turbo together with a custom-trained style LoRA to establish the overall mood, armor language, and gothic architectural direction.
Once a visual direction was selected, Nano Banana was used to generate structured character sheet references in order to stabilize proportions, silhouette consistency, and costume details across multiple scenes and camera angles.
The finalized character references were then brought back into ComfyUI and integrated into various cinematic environments to further explore lighting, framing, atmosphere, and visual storytelling.
The workflow focused heavily on:
rapid character ideation
iterative environment exploration
LoRA-assisted visual consistency
cinematic mood development
AI-assisted concept pipeline workflows
scalable character refinement inside ComfyUI
Tools used:
ComfyUI, Z-Image Turbo, Nano Banana, custom style LoRA
Ai To Gaussian Splat Loop Exploration
An experimental AI-to-3D workflow exploring cinematic looping animations generated from diffusion-based imagery and Gaussian Splat reconstruction.
The project started with a series of stylized AI-generated frames created using a custom-trained style LoRA to establish the visual identity, mood, and material language of the sequence.
Selected frames were then processed using Apple’s SHARP img2GaussianSplat reconstruction model to convert the generated imagery into navigable Gaussian Splat scenes.
The resulting splats were imported into Blender, where custom camera paths and seamless looping animations were developed to create fluid cinematic motion through the reconstructed environments.
The workflow explored:
AI-to-3D scene reconstruction
Gaussian Splat rendering pipelines
cinematic looping camera animation
stylized volumetric environments
hybrid diffusion + 3D workflows
experimental realtime-friendly scene generation
Tools used:
ComfyUI, custom style LoRA, Apple SHARP, Gaussian Splats, Blender, After Effects