AI
PYTHON
UX/UI
2025
Sketch to 3D
Creative Tech, AI, 3D
Sketch to 3D
Creative Tech, AI, 3D
OVERVIEW
A personal AI agent built into daily life — accessible via Telegram, with four distinct modes that handle emotional support, technical debugging, job search strategy, and everything in between. Not a demo. My actual workflow.
YEAR
2025
ROLE
Developer · UX/UI · Architect
12+
INTEGRATIONS
8
CRON JOBS
180
DAYS RUNNING
30+
TOOLS AVAILABLE
The Problem
A signature visual style is an asset — but it doesn't translate to 3D on its own. Studios, illustrators, and brands that want 3D versions of their work hit the same wall: hand-modeling every character is slow and expensive, and outsourced 3D drifts off-style. The exact look that makes the work valuable gets lost in translation.
The Solution
This pipeline locks the style first, then generates 3D from it. A LoRA trained on the target body of work keeps every output on-style. A single text prompt produces a style-consistent reference image (LoRA + Juggernaut-XL in ComfyUI), which feeds Meshy's image-to-3D to create a textured GLB. The model is committed to GitHub and an interactive viewer link comes back — all triggered from one Telegram message.
Why It Matters
On-brand 3D assets in minutes instead of days, straight from a prompt, with the style intact. And because the model retrains on any body of work, it isn't one artist's trick — it's a repeatable way to turn a visual identity into production-ready 3D. (Here, it's trained on my own sketchbook.)
LoRA Training
A custom LoRA adapter is trained on the curated dataset, fine-tuning Juggernaut-XL to generate images that are indistinguishable from the original illustration style. The heart of the pipeline.
Generate a 3D model from my sketchbook character 🎨
LIVE DEMO
This will hide itself!
Conversations
Four modes. Same agent. Pick one to see how the register shifts.
Glyph
The default register. Direct, observational, and slightly dry. Glyph knows your calendar, your projects, and what's still open. Short answers by default, longer only when asked.
“Calendar says 2pm. Decide what matters.”
AI
PYTHON
UX/UI
2025
Sketch to 3D
Creative Tech, AI, 3D
Sketch to 3D
Creative Tech, AI, 3D
OVERVIEW
A personal AI agent built into daily life — accessible via Telegram, with four distinct modes that handle emotional support, technical debugging, job search strategy, and everything in between. Not a demo. My actual workflow.
YEAR
2025
ROLE
Developer · UX/UI · Architect
12+
INTEGRATIONS
8
CRON JOBS
180
DAYS RUNNING
30+
TOOLS AVAILABLE
The Problem
A signature visual style is an asset — but it doesn't translate to 3D on its own. Studios, illustrators, and brands that want 3D versions of their work hit the same wall: hand-modeling every character is slow and expensive, and outsourced 3D drifts off-style. The exact look that makes the work valuable gets lost in translation.
The Solution
This pipeline locks the style first, then generates 3D from it. A LoRA trained on the target body of work keeps every output on-style. A single text prompt produces a style-consistent reference image (LoRA + Juggernaut-XL in ComfyUI), which feeds Meshy's image-to-3D to create a textured GLB. The model is committed to GitHub and an interactive viewer link comes back — all triggered from one Telegram message.
Why It Matters
On-brand 3D assets in minutes instead of days, straight from a prompt, with the style intact. And because the model retrains on any body of work, it isn't one artist's trick — it's a repeatable way to turn a visual identity into production-ready 3D. (Here, it's trained on my own sketchbook.)
LoRA Training
A custom LoRA adapter is trained on the curated dataset, fine-tuning Juggernaut-XL to generate images that are indistinguishable from the original illustration style. The heart of the pipeline.
Generate a 3D model from my sketchbook character 🎨
LIVE DEMO
This will hide itself!
Conversations
Four modes. Same agent. Pick one to see how the register shifts.
Glyph
The default register. Direct, observational, and slightly dry. Glyph knows your calendar, your projects, and what's still open. Short answers by default, longer only when asked.
“Calendar says 2pm. Decide what matters.”
AI
PYTHON
UX/UI
2025
Sketch to 3D
Creative Tech, AI, 3D
Sketch to 3D
Creative Tech, AI, 3D
OVERVIEW
A personal AI agent built into daily life — accessible via Telegram, with four distinct modes that handle emotional support, technical debugging, job search strategy, and everything in between. Not a demo. My actual workflow.
YEAR
2025
ROLE
Developer · UX/UI · Architect
12+
INTEGRATIONS
8
CRON JOBS
180
DAYS RUNNING
30+
TOOLS AVAILABLE
The Problem
A signature visual style is an asset — but it doesn't translate to 3D on its own. Studios, illustrators, and brands that want 3D versions of their work hit the same wall: hand-modeling every character is slow and expensive, and outsourced 3D drifts off-style. The exact look that makes the work valuable gets lost in translation.
The Solution
This pipeline locks the style first, then generates 3D from it. A LoRA trained on the target body of work keeps every output on-style. A single text prompt produces a style-consistent reference image (LoRA + Juggernaut-XL in ComfyUI), which feeds Meshy's image-to-3D to create a textured GLB. The model is committed to GitHub and an interactive viewer link comes back — all triggered from one Telegram message.
Why It Matters
On-brand 3D assets in minutes instead of days, straight from a prompt, with the style intact. And because the model retrains on any body of work, it isn't one artist's trick — it's a repeatable way to turn a visual identity into production-ready 3D. (Here, it's trained on my own sketchbook.)
LoRA Training
A custom LoRA adapter is trained on the curated dataset, fine-tuning Juggernaut-XL to generate images that are indistinguishable from the original illustration style. The heart of the pipeline.
Generate a 3D model from my sketchbook character 🎨
LIVE DEMO
This will hide itself!
Conversations
Four modes. Same agent. Pick one to see how the register shifts.
Glyph
The default register. Direct, observational, and slightly dry. Glyph knows your calendar, your projects, and what's still open. Short answers by default, longer only when asked.
“Calendar says 2pm. Decide what matters.”