Hermes: Theia — Codex Vision Skill

A Hermes Agent skill for image analysis and generation via the OpenAI Codex CLI.

What It Does

• Image analysis — Describe, QA, and understand images via Codex's vision capabilities
• Image generation — Generate new images using Codex's built-in image_gen tool (GPT-image-2)
• Identity-preserve generation — Use reference images to maintain consistent identity across generated portraits
• Batch generation — Generate multiple images from a shared reference with a prompt file

Requirements

| Requirement | Details | |---|---| | Codex CLI | npm install -g @openai/codex | | ChatGPT Plus | Required for image generation (OAuth, not API key) | | Auth | codex auth login completed — valid ~/.codex/auth.json |

Codex uses ChatGPT Plus OAuth tokens, not an API key. ~/.codex/auth.json has "auth_mode": "chatgpt" and "OPENAI_API_KEY": null. If tokens are invalidated, re-auth via codex auth login.

OpenRouter does not support image generation (404 on /v1/images/generations).

Quick Start

Analyze an Image

cd /path/to/images && codex exec \
  "Describe this image in detail" \
  --image photo.jpg \
  --skip-git-repo-check

Generate an Image (No Reference)

cd /path/to/output
codex exec --skip-git-repo-check -s workspace-write --add-dir /path/to/output \
  "Generate a photorealistic portrait. Save to /path/to/output/result.jpg"

Generate with Reference (Identity-Preserve)

codex exec --skip-git-repo-check -s workspace-write \
  --add-dir /path/to/references \
  -i primary-reference.jpg \
  --add-dir /path/to/output \
  "Generate a portrait. Reference identity from attached image. Save to /path/to/output/result.jpg"