AI Image Studio
Active
A dedicated AI image generation environment built on a separate LXC container — used to create all the artwork for this blog and other projects.
Since my mini PC has no GPU, heavy generation tasks are offloaded to RunPod cloud GPUs on demand, keeping costs low while still enabling high-quality output.
- Krita with AI Diffusion plugin for editing and generation
- ComfyUI for advanced workflow-based image pipelines
- noVNC browser access — no local install needed
- RunPod GPU on demand (RTX 3090 / 4090) for heavy tasks
- Juggernaut XL SDXL model for photorealistic output
- Used to generate all blog feature images and design assets
AI-Assisted Layout with Paper MCP
The studio is extended by a local MCP server that connects Claude directly to a live browser canvas. Pre-designed elements — images, badges, text blocks — can be arranged on a web page layout in real time. Agents reposition and resize elements autonomously based on design instructions, while I watch the result take shape live in the browser.
The workflow stays fully collaborative: I can move elements myself at any point, or give natural language instructions to redirect the agent. It bridges the gap between AI-generated assets and a finished, human-approved layout.
- Local MCP server connects Claude to a live browser canvas
- Agents position and resize design elements autonomously
- Real-time visual feedback — results visible immediately in the browser
- Manual override at any time — drag elements or give new instructions
- Bridges AI asset generation and final layout approval
Stack: LXC · Krita · ComfyUI · RunPod · Stable Diffusion XL · noVNC · Paper MCP