JCJair Carrillo

← Back to work

DevelopmentOffline Studio

Offline Studio

Personal ComfyUI runtime in Docker: models and identities on read-only binds, API bound to loopback only.

Need

Train and serve image workflows with personal LoRAs without copying weights, exposing the API to the network, or mixing personal identity with corporate applications.

What I built

A dedicated Docker component, an isolated bridge network, resource discovery, and a health check. End-to-end inference and containerized training are not validated yet.

Role

Design and implementation of the personal infra. Weights and dataset are not published.

Technical notes

API at http://127.0.0.1:8190, model downloads disabled, outputs in git-ignored artifacts. The Docker training recipe still needs validation.

Stack

  • Runtime: Docker, ComfyUI
  • AI: LoRA