Codex Imagen

Production image-gen toolkit on top of the Codex OAuth bridge — uses your ChatGPT subscription, no OPENAI_API_KEY needed.

Three interfaces over one engine: Python SDK, imagen CLI, MCP server.

---

Highlights

• 5 batch modes — single / parallel / variants / chain / branded-parallel (best for coherent website sections)
• 4 reasoning modes — raw (~20 s) → max (~3 min), auto-selected based on prompt shape
• Better transparency — green chroma-key + dominance despill, 54× cleaner on fur vs. magenta defaults
• Bundled AI skill — auto-installs into Claude Code / Codex / Claude Desktop on imagen setup
• Skill loader — pass any .md skill file, body merges into the prompt builder
• Codex labeled-spec prompts — 14-field structured prompts for predictable output
• MCP server — drop into 5 AI clients with one command

---

Install

One-liner (recommended):

# macOS / Linux
curl -fsSL https://raw.githubusercontent.com/VelmoAI/codex-imagen/main/install.sh | sh

# Windows PowerShell
iwr https://raw.githubusercontent.com/VelmoAI/codex-imagen/main/install.ps1 | iex

This installs uv (if missing) and runs imagen setup — registers the MCP server and the bundled skill in your AI clients.

Already have uv: uvx codex-imagen setup

pip: pip install codex-imagen && imagen setup

From GitHub (latest main): pip install git+https://github.com/VelmoAI/codex-imagen.git && imagen setup

---

60-second quick start

from codex_imagen import imagen

# One image
r = imagen(prompt="a brass diya oil lamp, editorial product shot")
print(r.images[0].path)

# 6 brand-consistent website sections in one call
r = imagen(
    prompt=["hero", "about", "features", "gallery", "pricing", "footer"],
    anchor="modern Indian restaurant, warm terracotta and brass palette, editorial photography",
    batch_mode="branded-parallel",
    parallel=3,
    output_dir="./out/restaurant-site",
)

CLI:

imagen "a brass diya oil lamp"
imagen "a brass diya oil lamp" --transparent       # RGBA PNG
imagen setup                                        # install MCP + skill
imagen status                                       # install state per client

---

Docs

• Batch modes — single, parallel, variants, chain, branded-parallel
• Reasoning modes — raw, medium, high, max — with timing
• Transparency — chroma-key pipeline, despill, complex subjects
• Skills — bundled skill + external skill loader
• MCP server — per-client setup (Claude Code, Desktop, Codex, Cursor)
• CLI reference — every subcommand and flag
• SDK reference — Python API, ImagenOptions, ImagenResult
• Why use this vs. native image_gen?
• Releasing

---

License

MIT, Copyright (c) 2026 wauwa. See LICENSE for the full text.