codex-image-generation

Dedicated Hermes skill for Codex OAuth + OpenAI hosted image_generation.

This repository is meant to be checked out directly at:

~/.hermes/skills/creative/codex-image-generation

It gives Hermes a git-managed image generation workflow with:

• direct Codex OAuth Responses transport;
• gpt-image-2 hosted image generation;
• fidelity-first prompt handling;
• direct, fidelity, and enhanced prompt modes;
• default --moderation low;
• exact visible text / language preservation policy;
• reference and edit image support;
• metadata and raw event capture;
• multiple candidate generation with --count.

Files

SKILL.md                  Hermes skill instructions
README.md                 Repository documentation
scripts/gen_image.py      Main CLI helper
scripts/test_gen_image.py Unit tests for payload, parsing, diagnostics

Requirements

• Hermes source at ~/.hermes/hermes-agent
• Hermes Codex OAuth credentials available
• Python environment with Hermes dependencies, including openai
• Optional: Pillow for compressing oversized reference images

Sanity check:

codex login status

The helper itself uses Hermes’ Codex OAuth helpers directly. It does not call the removed codex responses CLI subcommand.

Quick start

python ~/.hermes/skills/creative/codex-image-generation/scripts/gen_image.py \
  "A dreamy jellyfish library under the sea, painterly concept art" \
  --output ./jellyfish-library.png

Exact text / high-fidelity prompt:

python ~/.hermes/skills/creative/codex-image-generation/scripts/gen_image.py \
  "Create a clean event poster with the exact text 'FRI 11PM / HBC Rooftop', black background, silver typography, no extra text" \
  --prompt-mode fidelity \
  --output ./poster.png

Direct pass-through mode:

python ~/.hermes/skills/creative/codex-image-generation/scripts/gen_image.py \
  "Minimal red circle on a white background" \
  --prompt-mode direct \
  --output ./red-circle.png

Enhanced photoreal mode:

python ~/.hermes/skills/creative/codex-image-generation/scripts/gen_image.py \
  "A boutique hotel lobby in Seoul at night" \
  --prompt-mode enhanced \
  --enhancement-profile photoreal \
  --output ./hotel-lobby.png

Reference image:

python ~/.hermes/skills/creative/codex-image-generation/scripts/gen_image.py \
  "Create a luxury skincare ad using the attached bottle as the product reference" \
  --reference-image ./bottle.png \
  --output ./skincare-ad.png

Edit image:

python ~/.hermes/skills/creative/codex-image-generation/scripts/gen_image.py \
  "Replace the background with a rainy neon Tokyo street while keeping the subject intact" \
  --edit-image ./portrait.png \
  --action edit \
  --output ./portrait-tokyo.png

Multiple candidates:

python ~/.hermes/skills/creative/codex-image-generation/scripts/gen_image.py \
  "A cinematic sci-fi knight on a ruined moon" \