Designed and built a full-stack web application enabling AI-powered portrait generation and roleplay chat for user-defined characters, served at a custom domain
Engineered a PHP REST API (api.php) with auto-provisioning MySQL schema covering characters, gallery, jobs, and settings — all served via nginx reverse proxy
Integrated a self-hosted ComfyUI inference backend (CPU-only) via a custom async job queue with a PHP worker (worker.php), supporting txt2img, img2img, inpainting, and face-swap pipelines
Implemented a canvas-based inpainting tool with brush masking, GrowMask edge softening, and pixel-perfect compositing using ImageCompositeMasked + SetLatentNoiseMask
Added a ReActor face-swap pipeline using the ReActorFaceSwap ComfyUI custom node, with donor image upload and job retry support
Built a real-time job queue UI with live status polling, cache-busted gallery refresh, and direct image uploads
Implemented an Ollama-powered chat panel per character, with persistent chat history stored in MySQL and character-specific system prompts
Developed a pose/scene chip system with dynamic prompt assembly from character attributes, active LoRAs, trigger words, and scene presets
Managed full server infrastructure: systemd services, nginx config, Python venv, model and custom node installation
WebBetas