EDEN - Autonomous Creative AI Studio

The Challenge

The generative AI space is dominated by single-shot prompt-and-response tools: type a prompt, get an image, close the tab. Eden set out to build something more ambitious: a platform where creators could build persistent, autonomous AI agents with their own personalities, their own custom-trained visual models, and the ability to collaborate with humans and with each other on multi-step creative work. That meant solving problems the space doesn't have textbook answers for yet — training custom visual models on user-provided samples at reasonable cost, running a workflow ecosystem where community-contributed ComfyUI pipelines could be called autonomously by agents and earn their creators revenue, giving agents context, tools, and identity that persist across sessions, and building all of it on an open-source foundation so the platform stays extensible.

The Solution

I built the core experience at app.eden.art: a Next.js frontend and Fastify (Node.js) backend where creators build agents, train custom visual models, and generate multi-modal work. Behind it, a FastAPI (Python) orchestration service handles model calls, tool use, and agent execution — turning 'run this agent' into a chain of ComfyUI workflow calls that returns a finished output. A custom Flux LoRA training pipeline deployed on GCP lets users train visual models on their own sample images, so agent output reflects the creator's own style rather than a generic default. Community-contributed ComfyUI workflows become tools any agent can call, with contributors earning revenue when their workflows are used. On top of this, autonomous agents like Eve, Eden's flagship agent, can hold conversations and produce multi-step creative work — art, video, and stories — without a prompt at every step.

Key Features