Dockers-images

Documentation publique

Public portfolio repository for containerized AI workflows built and maintained by Unicorn Who Dev.

This repository documents practical work around reproducible GPU inference, model-serving environments, automated AI pipelines, preprocessing/post-processing stages, and deployment-oriented containers.

This is a portfolio and technical documentation repository. It does not expose production credentials, private endpoints, model secrets, proprietary datasets, or unrestricted paid GPU services.

What this demonstrates

• Docker-based AI deployment
• Reproducible GPU inference environments
• CUDA / PyTorch workloads
• Headless AI pipelines
• API-oriented model services
• Model and artifact volume separation
• Containerized preprocessing and post-processing
• Versioned runtime dependencies

Public Docker Hub

Docker Hub: https://hub.docker.com/u/charlibillabert

Public images are used as technical proof of containerized AI workflows. Individual containers may belong to active R&D projects and can have different runtime, model, licensing, and hardware requirements.

charlibillabert/mmg-pose-sheet-pipeline

A headless GPU pipeline created for a character-generation workflow.

Publicly inspectable image metadata shows a Linux/amd64 container built around:

The container packages the execution environment around a sketch-to-pose-sheet pipeline rather than relying on a manually configured workstation.

• CUDA 13.0
• PyTorch 2.11
• ComfyUI
• Diffusers
• version-pinned runtime components
• dedicated /models and /data/artifacts volumes
• API-oriented execution
• exposed service port 8788

Typical architecture

Model files and generated artifacts are kept separate from the container image when appropriate. This makes it possible to update runtime code independently from large model assets and persistent outputs.

Why containers

For AI workloads, Docker is used as part of the deployment architecture rather than only as a packaging step.

The goal is to make an inference pipeline:

• reproducible across machines,
• explicit about runtime dependencies,
• easier to deploy on remote GPU providers,
• easier to validate before accepting jobs,
• easier to version and roll back,
• and easier to integrate behind APIs or automated workflows.