Giggle Gen
The infrastructure behind AI image generation: connecting web requests to GPU processing, tracking jobs, and delivering generated media.
Project overview
Built a generation backend spanning Cloudflare Workers, Supabase, and Docker-based GPU workers running ComfyUI on local hardware and RunPod. The work brings together API development, database-backed queues, model distribution, and media delivery.
Engineering work
- Request handling and queues. Cloudflare Workers validate generation requests, verify Supabase access tokens, and create jobs with retry protection and per-user queue priorities.
- GPU processing. Containerised Python workers connect queued jobs to ComfyUI workflows, with model and LoRA configuration for local and RunPod deployments.
- Model distribution. Startup downloads from Cloudflare R2 allow model updates independently of container builds, with persistent storage options for reuse.
- Media delivery. An image-sharing Worker resolves image IDs through Supabase, streams files from a private R2 bucket, and caches responses.
- Video exploration. Separate ComfyUI containers explore text-to-video, image-to-video, and reference-to-video workflows.
Technologies
Python · JavaScript · SQL · Docker · Cloudflare Workers & R2 · Supabase · RunPod · ComfyUI · PowerShell
Discuss this project
CV summary
Built AI generation infrastructure integrating Cloudflare Workers, Supabase, and Dockerised GPU workers on RunPod and local hardware. Implemented authenticated job submission, queue prioritisation, model downloads from R2, and cached image delivery, with ComfyUI workflows for image generation and video experimentation.