Comfy Generation Manager

Comfy Generation Manager is the working interface around my local AI media pipeline. It turns folders of ComfyUI output into a searchable system with provenance, durable jobs, review decisions, repeatable workflows, agent tools, and a production Studio for assembling generated media into films.

This page is the current project reference. The repository contains setup and operational documentation; the related writing below preserves the design rationale, implementation journey, and lessons from building it.

At a glance

• Browse and filter images, video, and audio across multiple generation workers.
• Preserve prompts, workflow identity, inputs, lineage, ratings, and review data.
• Queue work durably so restarts do not erase intent or execution history.
• Launch curated workflow definitions through the UI or guarded agent tools.
• Organize selected generations into productions, scenes, shots, and takes.

Current technical shape

React + TypeScript interface
            |
            v
       FastAPI manager <---- agent tools
            |
      +-----+----------------+
      |                      |
      v                      v
durable SQLite         mounted media
      |
      v
ComfyUI workers + optional cloud generation

SQLite owns intent, state, review data, and queue history. Large media stays on disk. ComfyUI and optional cloud providers execute generations, while the manager remains the single path for durable submission and recovery.

Project boundary

The application is intentionally local-first and has no public hosted demo. The GitHub repository is the canonical source for the current implementation, installation instructions, runtime configuration, and operational commands.