← Selected WorkNiccolò Fenili Carravetta01 / 06

Stylized town environment generated from the scene workflow

Flagship case study2026

Unity × ComfyUICreative Pipeline

AI Creative Technology / Realtime Pipeline

Unity
ComfyUI
Flux
Generative AI
OneTrainer

01Overview

One modular bridge between realtime and generative workflows.

An experimental creative-technology pipeline connecting Unity with ComfyUI for realtime development, generative exploration and AI-assisted visual iteration.

Scene data, captures and authored inputs move from the realtime engine into reusable generative workflows, where they can be explored as visual alternatives and returned to the production context for further iteration.

The project brings texture generation, scene-driven imagery, VFX studies, consistent characters and custom LoRA training into one connected framework.

System: Unity ↔ ComfyUI
Focus: Reusable visual workflows
Discipline: Creative Technology

02System Overview

A bidirectional production loop.

Unity remains the realtime context. ComfyUI provides a modular processing layer for visual generation and controlled experimentation.

01UnityRealtime context
02Scene / input / captureAuthored signal
03ComfyUIModular workflow
04AI processingGeneration / variation
05Generated outputsReview / selection
06Realtime useReturn / iterate

03Realtime Texture Generation

From generated surface maps to a realtime material.

AI-assisted texture exploration is organized as a readable PBR sequence: surface colour, height and normal information, followed by the material in Unity and the recorded workflow.

Generated stone surface with cyan engraved symbols — 01
Surface colour / albedo study

Grayscale height map for the generated stone surface — 02
Height map

Purple-blue normal map for the generated stone surface — 03
Normal map

Generated stone material applied to realtime geometry — 04
Realtime material result

Unity ↔ ComfyUI workflow capture

04Unity Scene Capture → ComfyUI

The scene becomes the generative input.

Visual information captured from Unity is passed into ComfyUI to explore a new scene direction while preserving a clear relationship with the realtime source.

Simple realtime scene blockout in Unity — 01
Unity scene input

AI-assisted stylized interpretation of the Unity blockout — 02
Generated scene direction

Scene capture → ComfyUI workflow

05AI VFX Generation

Three visual studies, each shown in still and motion.

A sequence of AI-assisted VFX experiments explores how generated visual ideas can be evaluated inside a realtime presentation.

Cinematic energy-column VFX experiment in a stylized city — 01
VFX experiment / still

VFX experiment / motion

Twin sci-fi display VFX experiment — 03
VFX experiment / still

VFX experiment / motion

Stylized character VFX experiment in a realtime scene — 05
VFX experiment / still

VFX experiment / motion

06Consistent Character Generation

Identity held steady while the scene changes.

An advanced ComfyUI and Flux-based character workflow designed to preserve identity, outfit and visual language while allowing controlled variation in pose, expression, framing and environment.

Close character identity reference with a yellow hood — 01
Identity reference

Second close character identity reference with a yellow hood — 02
Identity reference

Full-body character identity and outfit reference — 03
Full-body identity reference

A fixed identity and outfit become the visual anchor for the following controlled variations.

Consistent character variation 01 with a yellow hood and dark outfit — 04
Controlled variation

Consistent character variation 02 with a yellow hood and dark outfit — 05
Controlled variation

Consistent character variation 03 with a yellow hood and dark outfit — 06
Controlled variation

Consistent character variation 04 with a yellow hood and dark outfit — 07
Controlled variation

Consistent character variation 05 with a yellow hood and dark outfit — 08
Controlled variation

Consistent character variation 06 with a yellow hood and dark outfit — 09
Controlled variation

Consistent character variation 07 with a yellow hood and dark outfit — 10
Controlled variation

Consistent character variation 08 with a yellow hood and dark outfit — 11
Controlled variation

Consistent character variation 09 with a yellow hood and dark outfit — 12
Controlled variation

Consistent character variation 10 with a yellow hood and dark outfit — 13
Controlled variation

07Custom LoRA Training

A reusable visual language beyond prompt-only generation.

A custom LoRA training experiment performed with OneTrainer, using curated visual references to explore consistent new props, characters and environments in a stylized high-fantasy game visual language.

Visual reference set

Custom LoRA visual variation 01 in a stylized high-fantasy visual language — 01
LoRA variation

Custom LoRA visual variation 02 in a stylized high-fantasy visual language — 02
LoRA variation

Custom LoRA visual variation 03 in a stylized high-fantasy visual language — 03
LoRA variation

Custom LoRA visual variation 04 in a stylized high-fantasy visual language — 04
LoRA variation

Custom LoRA visual variation 05 in a stylized high-fantasy visual language — 05
LoRA variation

Custom LoRA visual variation 06 in a stylized high-fantasy visual language — 06
LoRA variation

Custom LoRA visual variation 07 in a stylized high-fantasy visual language — 07
LoRA variation

Character & style variations

Custom LoRA visual variation 08 in a stylized high-fantasy visual language — 08
LoRA variation

Custom LoRA visual variation 09 in a stylized high-fantasy visual language — 09
LoRA variation

Custom LoRA visual variation 10 in a stylized high-fantasy visual language — 10
LoRA variation

Custom LoRA visual variation 11 in a stylized high-fantasy visual language — 11
LoRA variation

Custom LoRA visual variation 12 in a stylized high-fantasy visual language — 12
LoRA variation

Custom LoRA visual variation 13 in a stylized high-fantasy visual language — 13
LoRA variation

Custom LoRA visual variation 14 in a stylized high-fantasy visual language — 14
LoRA variation

Custom LoRA visual variation 15 in a stylized high-fantasy visual language — 15
LoRA variation

Custom LoRA visual variation 16 in a stylized high-fantasy visual language — 16
LoRA variation

Custom LoRA visual variation 17 in a stylized high-fantasy visual language — 17
LoRA variation

Custom LoRA visual variation 18 in a stylized high-fantasy visual language — 18
LoRA variation

Extended variations

Custom LoRA visual variation 19 in a stylized high-fantasy visual language — 19
LoRA variation

Custom LoRA visual variation 20 in a stylized high-fantasy visual language — 20
LoRA variation

Custom LoRA visual variation 21 in a stylized high-fantasy visual language — 21
LoRA variation

Custom LoRA visual variation 22 in a stylized high-fantasy visual language — 22
LoRA variation

Custom LoRA visual variation 23 in a stylized high-fantasy visual language — 23
LoRA variation

Custom LoRA visual variation 24 in a stylized high-fantasy visual language — 24
LoRA variation

Custom LoRA visual variation 25 in a stylized high-fantasy visual language — 25
LoRA variation

08Results / Capabilities

A compact map of the system’s scope.

The value of the project is the connected pipeline thinking: one modular framework for testing where realtime tools and generative workflows can meet.

01Unity ↔ ComfyUI pipeline thinking
02Creative Technology
03Realtime + generative AI integration
04AI-assisted PBR texture generation
05Scene-driven AI generation
06VFX experimentation
07Consistent character generation
08Flux-based workflows
09Custom LoRA training with OneTrainer
10Reusable AI production workflows

Next project · 02 / 06

The Last Layer2026 · AI CREATIVE TECHNOLOGY / AI-DRIVEN 2D GAME PROTOTYPE